EP1649405A2 - METHODE ET SYSTEME DE SELECTION DE CIBLES THERAPEUTIQUES PAR L'UTILISATION DE RESEAUX DYNAMIQUES D INTERACTIONS MOLECULA IRES - Google Patents
METHODE ET SYSTEME DE SELECTION DE CIBLES THERAPEUTIQUES PAR L'UTILISATION DE RESEAUX DYNAMIQUES D INTERACTIONS MOLECULA IRESInfo
- Publication number
- EP1649405A2 EP1649405A2 EP04786022A EP04786022A EP1649405A2 EP 1649405 A2 EP1649405 A2 EP 1649405A2 EP 04786022 A EP04786022 A EP 04786022A EP 04786022 A EP04786022 A EP 04786022A EP 1649405 A2 EP1649405 A2 EP 1649405A2
- Authority
- EP
- European Patent Office
- Prior art keywords
- molecules
- state
- graph
- vertices
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B5/00—ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B5/00—ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
- G16B5/30—Dynamic-time models
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B5/00—ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
- G16B5/10—Boolean models
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B5/00—ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
- G16B5/20—Probabilistic models
Definitions
- the present invention relates to the field of integrative analysis of molecular interactions in a biological system. It relates in particular to methods for obtaining and analyzing networks of biological molecular interactions making it possible, from obtaining experimental data, to identify and describe the functioning of these interactions at the same time (i) between molecules interacting two by two, (ii) at the level of the results of the interactions exerted on a given molecule, and (iii) at the level of the whole network of interactions considered. Even more particularly, this method of analysis makes it possible, once the functioning of these interactions has been described, to predict the consequences, over the entire network of molecular interactions considered, of activations or inhibitions of the molecules forming this network. It thus makes it possible in particular to identify potential therapeutic targets, to understand the mechanisms of action of xenobiotics.
- the approach which is increasingly used to develop new pharmacological treatments is another approach, called “physiological” or “comprehensive”, which consists in exploring and understanding the pathophysiological mechanisms, and in particular the mechanisms molecular pathophysiology, leading to the pathology concerned, in order to define the molecules of the organism to be treated, which will be the target molecules (or “therapeutic targets") of chemical treatments.
- the identification of these target molecules then makes it possible, in a second step, to carry out sieves of potential therapeutic molecules of synthesis in order to identify those which will directly modify the biological activity of these therapeutic targets, or even to perform oriented syntheses of such therapeutic molecules when the spatial structure of the target molecules is known.
- Integrative biology methods aim to analyze the role of molecules present in the organism to be treated, taking into account (and therefore integrating) in this analysis the other molecules with which they interact. Their objective is therefore to make it possible to obtain models of the cascades (or networks) of molecular interactions of living organisms, in particular those involved in pathological processes. In the context of the selection of therapeutic targets, such models aim to be applied to select these targets. More precisely, such an application must make it possible to predict the consequences of the actions of activation or inhibition of the molecules of the network, in order to identify those which will have a therapeutic effect. It is only conceivable to make such predictions on a large scale and in a sufficiently reliable manner if the model allows systematic simulations of the effects of inhibition or activation actions of the molecules of the cascade.
- the modeling methods proposed today are, on the one hand, methods producing static models and, on the other hand, methods producing dynamic models.
- the modeling methods producing static models consist in constructing static graphs representing cascades of interactions of biological molecules from data in the scientific literature (publications in journals, analysis of expression profiles of molecules, taking into account sequence data, etc.).
- the resulting graph can be represented in the form of a diagram, most often in two dimensions, whose nodes (or vertices) of the graph are the molecules, and where these nodes are connected by lines or arrows (or arcs, or vertices of the graph) representing the interactions between molecules.
- Examples of static graphs are those constructed in various public databases such as for example the KEGG database (M. Kanehisa and S. Goto: KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Research, 28 (1): 27- 30, 2000).
- Diffusion the diffusion of molecules in the biological system studied or outside the biological system studied (for example a cell) is also taken into account, as equivalent to a synthesis or a degradation (respectively) within the system.
- Variables used all the existing methods define the variables as being the rate, or the concentration, or the total quantity, of the molecules, noted here Xj for the molecule i, and not its proportion of variation compared to a standard state x i0 .
- Continuous functions for a continuous function, the variables change continuously (as is the case in real biological systems) and not discrete.
- Deterministic model once the model has been calculated, the network can only pass from one state to another by a single path (unique sequence of intermediate states). The fact that a model is deterministic makes it possible to obtain a linear growth in the quantity of calculations during simulations. Conversely, in non-deterministic models, the amount of computation required during simulations tends to increase exponentially with the size of the network, which can lead to an impossibility of implementation for large networks.
- Level A knowledge of the existence in itself of molecular interactions, and at least part of their orientations and part of the effects of interactions (activation / inhibition or synthesis / degradation). Only level A knowledge is widely available to date. Therefore, only a method requiring only level A knowledge can be applied to wide area networks.
- Level B level A with all the directions of the interactions and all the effects of the interactions.
- Level C extensive functional knowledge of the network, i.e. level B plus other data such as: chemical reaction rate constants, description of threshold effects, description of allosteric effects, etc. To date, whatever the living organism considered, for most of the molecules in the molecular interaction networks, level C knowledge is not available. A detailed functional description of the biological network is necessary for the implementation of the method when level C knowledge is required.
- any method requiring this type of knowledge for its implementation can only be applied to very small well-studied and known networks (a few dozen molecules at maximum) and is in fact unsuitable for its application to large networks (more than 100 to 150 molecules).
- Level A linear growth with the size of the network (in number of molecules) of the required amount of computation. This corresponds to the possibility of implementation on a standard power server (general public). The methods implementing calculations, the quantity of which increases linearly with the size of the network, can be applied to large networks (provided that there is no other limit to this application). Level B: growth in the amount of intermediate computation between cases A and C. The methods implementing computations whose quantity increases in an intermediate manner between A and C are theoretically applicable to large networks but at a high or very high cost (and subject to not presenting any other limit to this application).
- Level C exponential growth with the size of the network (in number of molecules) of the amount of calculation required. Any method implementing calculations, the quantity of which increases exponentially with the size of the network, requires very high computing power. For example, some applications of Bayesian networks require around 30 minutes of computing time on a server equipped with a 1.2 Giga Hertz processor for a network of 8 molecules: on a network of 32 molecules, the time to calculation on the same computer would in this case be more than a year and a half. In practice, even with today's most powerful computers, the methods exhibiting an exponential growth in computing time are not applicable to large or very large networks (a few thousand to a few tens of thousands of molecules and more; some of them are not applicable even to networks of a few hundred molecules). (3) Maximum size of network implemented: this is the maximum size of the networks on which the method has been successfully implemented to date.
- the present invention aims to provide a method for obtaining dynamic models of molecular interaction networks in a biological system, which make this type of application possible. For the sake of clarity of this text, a certain number of terms are defined below.
- This increase (or decrease, respectively) in biological activity can correspond either to an increase (or a decrease, respectively) in the number of molecules of a given type present in the biological system analyzed, each keeping the same activity (or biological function, either to an increase (or a decrease, respectively) in the activity of molecules of a given type, their number remaining constant, or to a combination of these two mechanisms, or to the result of these two mechanisms.
- Activation (or inhibition, respectively) can also be: the consequence of an increase (or a decrease, respectively) in the number of molecules associated with a decrease - (or an increase, respectively) in their biological activity , if the overall result is an overall increase (or an overall decrease, respectively) in activity, and vice versa.
- Activation can be non-zero or zero depending on the molecules considered and the biological system considered. It can be variable over time. The fact that certain interactions of the network of molecular interactions considered correspond to zero activation (or inhibition, respectively) is only a particular case of the field of the invention.
- the biological activity of a biological molecule (s) considered corresponds to any capacity of the molecule (s) considered to have a chemical and / or physical interaction with any other molecule. another type (or with another molecule of the same type). This chemical and / or physical interaction may or may not result in the acquisition (or loss) by one of the interacting molecules of capacities to have a chemical and / or physical interaction with any other molecule of another type (or with another molecule of the same type).
- Chemical interactions are any interaction between two (or more) molecules causing a chemical reaction (which can be represented by a modification of the formula of a molecule, or the synthesis, or the degradation of a molecule).
- Physical interactions are any interaction between two (or more) molecules causing the formation of a stable or unstable complex between these molecules.
- Examples of biological activities of molecules and corresponding molecular interactions are (not exclusively): the activity of activating the transcription of a given gene (molecular interaction: protein (transcription factor) - DNA), the activity of implementing a chemical reaction (molecular interaction: protein (enzyme) - molecule (substrate), allowing the transformation of the molecule-substrate into molecule-product of the chemical reaction), the activity of formation of '' a protein molecular complex itself having such or such biological activity (molecular interaction: protein (complex subunit) - protein (complex subunit)), etc.
- biological molecule it is understood here any molecule, whatever its complexity, present in the biological system considered.
- biological system it is understood here any living organism, whether it is prokaryotic or eukaryotic, and whether it is unicellular or pluricellular, and that the biological system corresponds to this organism as a whole or to a part of this organism.
- biological system it is understood here any living organism, whether it is prokaryotic or eukaryotic, and whether it is unicellular or pluricellular, and that the biological system corresponds to this organism as a whole or to a part of this organism.
- Whole organisms - A cell (eukaryotic or prokaryotic) as a whole. - A set of cells interacting directly or indirectly with one another, or not interacting with each other: 0 all of the cells in culture in a petri dish; 0 all of the cells forming an organ or part of that organ: the tonsil nucleus of a mammalian brain. - A multicellular living being. - The different examples plus their environment. Part of an organism: - An organelle of a cell, such as a mitochondrion.
- a set of molecules participating in a given biological function such as a set of molecules participating in cellular respiration, or a set of molecules participating in cell death, whether this set of molecules is made up of all the molecules participating in said biological function where only a part of them.
- the set of molecules forming the network of molecular interactions as described in the form of a static graph in FIG. 2 is an example of a biological system.
- Many static graphs are, for example, available in the public KEGG database (M. Kanehisa and S. Goto: KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Research, 28 (1): 27-30, 2000). Every biological system is made up of molecules, these molecules interacting with each other in a more or less stable and variable manner over time and the effects of the environment of this system on the biological system itself.
- apoptosis is the result of the interaction of multiple molecules (hormones, proteins, second messengers, etc.) which, for some of them, have physical interactions or chemical more or less stable over time.
- network of molecular interactions is understood here the set of molecules analyzed by the method of the invention associated with the set (or part of this set) of their possible biological interactions.
- the network can include all the molecules of the biological system concerned, or only a part of these molecules.
- the network can be represented visually in the form of a graph (as an example is given in the description below). It is this type of visual representation that is behind the use of the term "network". However, such a representation is not a prerequisite of the invention.
- the network can also be represented by a table (or a matrix) for which each row corresponds to one of the molecules of the network and whose columns correspond to the characteristics of the possible biological interactions of these molecules (or of a part of these interactions or their characteristics).
- a graph is here a representation of the network of molecular interactions in the form of a graph whose vertices (or nodes) correspond to the molecules of the network of molecular interactions represented and whose edges (or arcs) connecting the vertices correspond to the interactions molecules of the network of molecular interactions represented.
- vertices or nodes
- edges or arcs
- variable associated with a vertex of the graph it is understood here a quantitative variable in the mathematical sense of the term, which can take numerical values, and whose value at a given state of the graph represents the state of the corresponding vertex with regard to a quantity relating to a molecule of the biological system considered.
- this quantity can be a level of expression of a gene expressed in the biological system (for example, the abundance of messenger RNA, measurable in particular by the microarray technique), a level of abundance of a protein, a level of activity of a protein, a level of abundance of a metabolite, etc., provided that the quantity considered is measurable experimentally, by a direct means or not.
- a state of a graph is a graph for which a numerical value is given for each variable (associated with each vertex). The case where a nonzero numerical value is given only for a part of the variables (and associated with the corresponding vertices), another part of the variables (associated with other vertices) being zero, is only a particular case state of the graph.
- a state of the given graph is a representation of a real or simulated state of the corresponding network of molecular interactions, and by extension a representation of a real or simulated state of the corresponding biological system.
- giving a variable associated with a vertex of the graph the value zero can correspond to a representation of the situation where the molecule corresponding to this vertex is not present in the network of interactions (which does not mean that it is not present in the biological system), or else in the situation where its biological activity is zero.
- the fact of giving a null value to a certain number of variables therefore corresponds to considering that at a given time these do not interact with the rest of the network, but their value can become non-zero at another time following a modification of the network status. Giving a variable a null value therefore does not necessarily amount to excluding the corresponding vertex from the network.
- the invention also relates to a computer system for obtaining a dynamic model of a network of molecular interactions in a biological system, and the analysis of these molecular interactions when a stimulus is applied to the dynamic model, comprising at least at least one central data processing unit connected to at least one quantitative experimental database, the computer system comprising: A) a module for constructing a static graph, the vertices of which represent biological molecules and the arcs of which represent physico-chemical interactions existing between these molecules, each vertex being associated with a quantitative variable measured experimentally and each arc of the graph being associated with a mathematical relationship; and B) a learning module for calculating the parameters of each relation from quantitative experimental data concerning the vertices of the graph, by implementing learning techniques using gradient descent used for the configuration of networks.
- the computer system according to the invention can also comprise: C) a simulation module for performing several iterative simulation procedures consisting in imposing a stimulus on a graph state measured experimentally and chosen as "state to modify", the stimulus modifying the value of one or more of the quantitative variables associated with the vertices of the graph, thus constituting a starting state of the simulation from which a propagation calculation is performed within the graph, to obtain a "final state of the graph ”; and D) an iteration module for modifying the stimulus.
- the computer system according to the invention may further comprise: E) a module for calculating proximity between the “final state of a graph” and the “state to be modified”, or between the “final state of a graph ”and a desired state, and hierarchy of vertices and stimuli imposed on the vertices of the graph, the hierarchical vertices corresponding to classified therapeutic targets.
- the computer system forms a tool for analyzing biological experimental data, and in particular a tool for hierarchy of biological molecules vis-à-vis a biological problem.
- a first aspect of the invention is a method for obtaining a dynamic model of a network of molecular interactions.
- the real sign of the term (ii) is determined by the result of the calculation of its parameter (s). This term (ii) is of opposite sign to term (i) once the parameters have been calculated, but this does not necessarily appear in its mathematical formulation, where the sign of the parameter (s) is not specified a priori shareholder (s).
- each quantitative variable associated with a vertex represents the relative variation of the quantity of the molecule corresponding to said vertex, compared to a standard state of the biological system.
- the "quantity of the molecule associated with a vertex" can relate to any aspect that can be measured directly or indirectly, this molecule, whether it is its concentration, its activity, its level of expression , etc.
- said standard state is preferably a stable state of the biological system, in which the quantity of each molecule associated with a vertex of the graph is measurable experimentally.
- this standard state can correspond to a given physiological state (for example healthy or sick) actually observable, or to an artificial state of the system, for example in the state a pool of several biological samples taken under different experimental conditions.
- the relative variations in quantity of the molecules of the network are therefore represented in the form of variables dependent on the relative variations in quantity of the molecules interacting on them (ie in interaction with them and upstream in the network in terms of propagation of activations / inhibitions).
- the definition of the variables corresponds directly to the experimental measurements available: in fact, in most molecular biology technologies (including screenings of messenger RNA expression), the absolute quantity of molecules present in the biological system of interest .- is not measured (nor measurable); only the proportion of their variation compared to a reference state is measurable.
- the terms inertial (i) and return to the initial state (ii) make it possible to calculate the values of Xi and the variations of values of Xi over time as a function of the values of Xj (1- »n) and of the variations of values of X j (1-» n) over time.
- inertial term is meant: - resistance to change, resistance in particular initial, and - a time to arrive at the maximum variation, which makes it possible to account for the complexities of propagations in the network.
- the purpose of the inertial term (i) is to make it possible to integrate a resistance of the variables to change and a time difference between the modifications of the variables upstream and downstream of the network. It introduces in particular: - the integration of the time factor - the taking into account of the propagation speed differences within the network as a function of the sub-circuits - the taking into account of the temporal delays consecutive to the influences of the feedback loops on the propagation in the network, and - it makes it possible to calculate the kinetics of molecular interactions within the network directly from experimental data, without prior knowledge of the speed constants of these kinetics, and without making a priori on any other parameters.
- This inertial term (i) tends towards a finite limit, which makes it possible to avoid significant divergences during simulations (improvement of their reliability): this avoids the risk of divergence (or "explosion") of the values of the linked variables to iterative propagations in feedback loops or during simulations involving extended times.
- divergence or "explosion"
- the formulation of this inertial term is preferably not very restrictive, and makes it possible to account for multiple forms of relations.
- any initial variation in the effect of term (ii) on the calculated value of Xj can therefore be considered as consecutive to a prior variation in the effect of term (i) on the calculated value of Xj.
- Xj can only present a variation if, at at least a given time, the variation of Xj at the following time calculated by the term (ii) is less in absolute value than the variation of Xj at the following time calculated by the term (i).
- Xj can only present a variation, from a stable state, if, over at least a given time space, the variation of the calculated value of the term (ii) is less in absolute value than the change in the calculated value of term (i).
- This characteristic is inherent in the fact that the term (i) is expressed as a function of X j (1 ⁇ n) while the term (ii) is expressed as a function of Xj.
- the variation of term (ii) is initially lower in absolute value than the variation of term (i).
- the effect of term (ii) on the variation of Xj may or may not become greater than the effect of term (i) on the variation of Xj. If this is the case, Xj will tend to return to its initial value.
- the methods for obtaining a dynamic model of a network of molecular interactions in a biological system comprise a second step (step B), in which the parameters of the associated relationships are calculated. to each of the arcs of the graph, from quantitative experimental data concerning the vertices of the graph. This calculation is preferably carried out by the use of learning techniques.
- a dynamic graph entirely deterministic, consisting of the static graph at the edges of which are now associated mathematical relationships whose parameters have all been defined numerically.
- This calculation step can be performed by the use of learning procedures used for the configuration of artificial intelligence networks, for example those developed in computer science in "neural network” methods (including recurrent neural networks) by "simple” gradient descent (taking as a basis for calculation the pairs of data (Xj, X j ) provided by the experimental data independently of each other), or by gradient descent in time (where these couples are not considered independent).
- the data pairs (Xj, X j ) provided by the experimental data are defined as follows: either i a molecule of the network, represented by the commit i, and either j any molecule of the network interacting on i, represented by the vertex j.
- Xj and X j are the variables associated with vertices i and j, respectively.
- experimental measurements of the values of Xj and X j under given experimental conditions and at given experimental times make it possible to obtain numerical values of Xj and X j .
- a pair of experimental data (Xj, X j ) corresponds to the measured values of Xj and X j at a given experimental state (same time, same experimental condition).
- the experimental data used to carry out step B) mentioned above have the following characteristics: Nature of the experimental data.
- Those data are quantitative data concerning the molecules (corresponding to the vertices of the graph) and are for example levels of expression of genes expressed in the biological system (by measuring the abundance of messenger RNA, for example by the microarray technique DNA levels) and / or protein abundance levels and / or protein activity levels and / or metabolite abundance levels.
- these data can be expressed in the form of a proportion of variation in quantity compared to a reference situation (standard state).
- indexing of experimental data can be automatically indexed to the graph.
- the role of this indexing is to link each experimental data to the corresponding biological object of the graph (vertex of the graph, or stop of the graph for the data pairs (Xj, X j ), so that these two types of information (experimental data and graph) can be used together when implementing the parameter calculation system.
- Many commercial or free database systems allow this indexing to be created without any particular technical difficulty for those skilled in the art of biology or bioinformatics (Oracle commercial databases, Microsoft SQL server, FileMaker , open access databases: postgreSQL).
- this indexing step may not be necessary per se.
- the experimental data for the values of the pairs (Xj, X j ) are in the form of expression kinetics.
- expression kinetics is understood here a set of series of experimental data ordered in time, each series of data corresponding to a set of values of couples (Xj, X j ) measured experimentally at a given time.
- Each series of data can concern either the set of vertices of the graph, or only a subset of these vertices.
- the different times correspond to successive times during the observation of a biological process using the biological system modeled by the graph, whether this process is natural or artificially induced in the laboratory.
- Such kinetics preferably comprises at least three successive times, and, to improve the quality of the calculation of the parameters, more than three times.
- step B) of the methods of the invention preferably takes account of the following principles:
- the graph is considered to be in a stable state of reference at a given time, this stable state being measurable experimentally.
- the stable reference state in question corresponds to an existing and measurable state of the biological system studied, which can be considered to be stable over time with respect to the modeled biological process.
- the standard state which is arbitrarily defined by the experimental biologist, is used to carry out quantitative experimental measurements.
- the stable reference state corresponds to a real state of the modeled system (Le., Not artificial), and serves as a reference for the calculation of the parameters of the model. It is considered a state of the system where the activation and inhibition processes at within the network are balanced, or have weak oscillations around a theoretical state of equilibrium. It represents the state towards which the system generally tends to return during simulations. It can be the same, or different, from the standard state.
- the stable reference state is directly measurable experimentally as soon as the biological problem studied makes it possible to define a reference state of the biological system.
- a cell culture whose number of cells has reached a plateau (absence of cell divisions) and in a stable culture medium, before any induction of stimulus, or a healthy adult animal before any induction of pathological process can be considered as stable reference states.
- the cascades of molecular interactions brought into play by the stimulus whose consequences one seeks to model are not activated beyond the homeostatic processes.
- the cascades brought into play by the pathological process to be modeled are also not implemented: the reference state is stable with respect to the modeled biological process.
- the stable state does not necessarily have to be the initial state of the biological system within the framework of the biological process studied.
- the healthy state can be considered as an initial stable state of reference if we study the installation of a pathological process starting from this healthy state.
- the measurement of the Xj of all the vertices of the graph in this state is used, in the calculation of the parameters, as a stable reference of the graph, in particular for the error minimization procedure.
- the stable state is mathematically defined by the vector of the set of experimental values of the variables of each vertex measured in the corresponding biological state (measurements made for all the vertices of the graph).
- the standard state for the measurements is the stable state.
- the Xj are close to 1 in general (if the standard measurement state is time t 0 ).
- V i, X, 1, and to introduce (in the sense of "adding") this vector of Xj at the initial kinetic time without having measured it.
- the standard state is considered stable, arbitrarily. This is often possible if the standard state does not correspond to a pool of different biological tissues.
- the experimental data are measured during kinetics (see above). In the case where the biological process of interest is studied during the transition from a state initial stable to a final stable state, and where experimental measurements are made at these two states and at intermediate times, two stable states are defined: the initial state and the final state of the kinetics of the biological process studied. However, having experimental measurements corresponding to two stable states is not a prerequisite for implementing the invention.
- the calculation of the parameters of the relationships (Xj, X j ) is carried out by the implementation of learning techniques used for the configuration of artificial intelligence networks (such as those implemented for neural networks), from quantitative experimental data concerning the variables of the graph.
- this calculation can use algorithms for digital resolution of propagation or back propagation with calculation of the error.
- a second aspect of the present invention relates to a method for analyzing a network of molecular interactions in a biological system, comprising the following steps: A ′) use of a dynamic model of the network of molecular interactions, said model being likely to be obtained by a process described above, and constructed from a static graph whose vertices represent biological molecules of the biological system and the edges represent physicochemical interactions between these molecules, and from experimental data regarding the levels or activities of these biological molecules.
- the propagation calculation within the graph can be performed for a number of time steps such that the duration of the simulation does not exceed the duration of the biological process to be simulated defined in step C). However, it is also possible to let the simulation continue beyond the duration of the biological process to be simulated defined in step C), for example if we are looking to see if the network will eventually find a new stable state (equilibrium state) and if we do not know a priori how long it will take. It is important to note that the duration of the simulation defined in step C) can be longer than that of the experimental kinetics used for the calculation of the parameters (or shorter).
- steps C), D) a) and D) b) above are carried out, using (without reconstructing) a dynamic model of the network of molecular interactions chosen, said model being capable of being obtained by a method such as the methods for obtaining dynamic models of networks of molecular interactions described above.
- Another particularly important aspect of the present invention is a method for selecting therapeutic targets implementing a dynamic model of a network of molecular interactions in a biological system, by implementing a computer system, comprising the steps and following characteristics:
- D) several iterative simulation procedures are carried out, each comprising the following steps: a) a stimulus is imposed on the state to be modified, that is
- step D) b) can be continued beyond the duration specified in step C)
- Step A ') of the above methods can be carried out in the same way as steps A) and B) of the methods for obtaining dynamic models of interaction networks described above.
- step C) can be carried out by taking account of the following elements, when the case permits:
- Time step The duration of the biological process to be simulated is divided into a series of time steps, regularly spaced or not; the time steps are defined so as to be preferably smaller than the real experimental durations separating the series of quantitative experimental data used for the computation of the parameters of the relations.
- the definition of these time steps is made necessary by the fact that any computer process of dynamic simulation consists in calculating states at discrete times, making the discretization of the time necessary. We therefore obtain a series of consecutive times, on which the simulation will be performed.
- the first beat in the series is called the initial beat. This initial time corresponds to the starting state of the graph, defined below.
- the state to be modified and the state to be achieved are defined as follows: We simulate actions on the vertices of the graph from a state to modify the graph identical or similar to its state as observed experimentally in the pathological condition (for example by screening the expression of messenger RNA on chips DNA from pathological tissue).
- the state to be reached is defined as being a state close to a reference non-pathological state (as also measured by the experimental observation of the non-pathological condition, for example by screening for expression of the messenger RNAs on microchips). DNA from healthy tissue).
- the simulation process then consists in identifying the vertices, and the stimuli on these vertices, which, starting from the state to be modified (the pathological state), best allow the graph to evolve (in part or entirely) towards a state close to the state to be reached (non-pathological state).
- the state to be modified is defined as above, and the state to be reached is defined as the state, or a state close, to that obtained experimentally during the administration of this treatment (as measured for example by screening the expression of messenger RNAs on DNA chips from pathological tissues which have been subjected to the treatment concerned).
- the simulation process then consists in identifying the vertices, and the stimuli on these vertices, which, starting from the state to be modified (the pathological state), best allow the graph to evolve (in part or entirely) towards a state close to the state to be reached (pathological state under treatment).
- This particular implementation can also be carried out by defining the state to be modified as any possible state G of the biological system studied (for example the healthy state), and the state to be reached as the state obtained after administration of the treatment concerned with the biological system in state G.
- step D) is carried out by considering the following elements:
- Stimulus A stimulus is imposed on the state to be modified. This stimulus is exerted in the form of the variation of the value of one or more variable (s) of the graph (corresponding to one or more vertex (s)), that is to say an increase or d 'a decrease in this or these value (s), depending on the desired simulation. The values of all other variables remain unchanged. We therefore obtain a new state of the graph, which is
- starting state of the simulation.
- the starting state and the state to be modified therefore only differ by the value of the modified variable (s), all the values of all the other variables being identical
- This state is defined as corresponding to the first step of the simulation
- the stimuli relate each time to a single vertex.
- Propagation From the starting state of the simulation, a propagation calculation is performed within the network. This propagation consists in calculating the new values of all the variables in the next time step, resulting in a new state of the graph, and in starting again the calculation starting from this new state for the next time step, and so on. This propagation is prolonged during the number of time steps (thus the biological duration) defined by the experimenter according to the biological question posed. It can possibly be extended until the appearance of a new stable state of the graph (a new state of equilibrium), or be stopped before. At the end of this simulation, a new state ("final state”) of the graph is obtained.
- the previous process is repeated with a new stimulus, relating to one or more other vertex (es) of the graph, or possibly relating to the same vertex (s) of the graph with the imposition a new value to the variable (s).
- step E) is a hierarchy of the vertices, and of the stimuli imposed on these vertices.
- This classification therefore corresponds to the classification of the vertices, from that on which a stimulus is most likely to result in the desired state from the state to be modified, to that on which a stimulus is least likely to have this effect.
- Each proximity corresponds in fact to one and only one vertex and one and only one stimulation value on this vertex. If the desired effect is the improvement of a pathological state, this classification is that of potential therapeutic targets, most likely to least likely.
- step D) c) the proximity of each final state obtained in step D) b) can be calculated either with respect to the state to be modified chosen in step C), or with respect to another state, measured experimentally or determined arbitrarily, and defined as P "state to reach", which can be, for example, a healthy state. It can be the reference state defined above.
- step E) consists in classifying all the bytes (vertex (s) of the graph - stimulus) in hierarchical order (ascending or descending) corresponding directly to the hierarchical order (ascending or descending, respectively) of the proximities associated with them.
- the vertices of the graph directly correspond to the molecules of the biological network, which are therefore hierarchized in fact.
- This hierarchy poses no technical problem to those skilled in the art, the proximities being positive numerical values which can be directly hierarchized from the largest to the smallest, or vice versa.
- the result of this classification can advantageously be produced in the form of a list or table, or in any other type of format, and / or stored in a database for later use.
- the hierarchical classification of the molecules of the biological network corresponds directly to the hierarchical classification of these molecules considered as therapeutic targets.
- the invention therefore makes it possible to obtain a classification of potential therapeutic targets hierarchized according to objective statistical criteria, as a function of experimental data (measurements of Xj) and of functional knowledge of the network (existence of molecular interactions).
- the best potential therapeutic targets are considered to be those corresponding to the best proximity to the state to be reached.
- this invention makes it possible not only to identify therapeutic target molecules, but also to predict the direction of the therapeutic action which it will be necessary to apply to these molecules (activation or inhibition).
- the therapeutic targets are therefore selected from data concerning all of the molecules studied, and not only those specifically concerning target molecules, since the criterion used for prioritization depends on the evolution of the graph as a whole, therefore on the set of experimental measures of expression and / or activation of all the molecules represented in the graph , and not the simple evolution of experimental measures of expression and / or activation of only target molecules. It is therefore indeed an integrative method responding to current needs as defined above, in particular as regards diseases with multi-factorial determinism, clearly bringing progress compared to the methods of selection of existing therapeutic targets.
- the target identification method described above has the following advantageous characteristics: -
- the calculations are based on a non-probabilistic method, which eliminates any limitation in terms of calculation time, unlike the methods of stochastic equations and Bayesian networks .
- the invention integrates quantitative experimental data, which differentiates it from qualitative methods (Boolean networks, generalized logical formalisms, formalisms based on rules), makes it possible to avoid constraints and hypotheses on the functioning of the network, and makes it possible to increase the reliability of simulations.
- the fact of defining the variables as similar to the actually measurable experimental data makes it possible to calculate the parameters of the relationships in an optimal manner (without having to extrapolate the values of the variables).
- a first hierarchical classification of the vertices, considered individually, is obtained by performing steps A) to E) by imposing, for each of the simulations of step D), stimuli that relate to a single summit; a step D2) is then carried out, corresponding to step D) in which the stimuli imposed on each simulation are exercised on two vertices, either by testing all the combinations of two possible vertices, or by limiting these calculations to the combinations of two vertices among a certain number of the vertices best classified in step E).
- a step E2) of hierarchical classification of the associations of two vertices on which stimuli are most likely to have the desired effect is carried out from the set of statistical proximities calculated in step D2).
- steps D) and E) can be repeated iteratively, each time increasing the number of vertices on which the stimuli are exerted.
- the method can include a step D3) following step E2) and corresponding to step D) in which the stimuli imposed on each simulation are exerted on three vertices, either by testing all the combinations of three possible vertices, or by limiting these calculations to the combinations of three vertices chosen from a number of the highest classified vertices in step E) and combinations of two best classified vertices in step E2), said step D3) being followed by a step E3) of hierarchical classification of the associations of three vertices on which stimuli are most likely to have the desired effect.
- Steps D4) and E4), with stimuli on 4 vertices can then be added, and so on.
- These methods of selecting therapeutic targets preferably include a final step of statistical classification of the proximities of graphs of all the simulations carried out, integrating all the classifications previously obtained.
- the stimuli exerted on these different vertices can be applied simultaneously or not, that is to say that the simulation can take account of a time difference between the stimuli exerted on different vertices.
- the relationship between Xj and Xj is established, for at least part of the physicochemical interactions between the molecules of the network, by an inertial relationship arising from that of the harmonic oscillator in physics , associated with a sufficiently large damping factor to limit the oscillation to a single cycle.
- Wj j .X j mi. (D 2 Xi / dt 2 ) + 2 . ⁇ . (DXi / dt) + ⁇ y 2 .Xj, in which mi. (d 2 Xi / dt 2 ) + Guy 2 .Xj corresponds to the inertial term (i), 2 . ⁇ .
- (dXj / dt) corresponds to the term of return to the initial state ( ii)
- Xi is a variable associated with the molecule i
- dXi / dt is the derivative of Xj as a function of time d 2 X
- / dt 2 is the second derivative of Xj as a function of time
- X j is a variable associated with the molecule j
- ⁇ ii represents the inertia of i
- ⁇ y governs the return to the equilibrium state of Xj
- the pulsation ⁇ y corresponds to the response time of Xj to the variation of X j
- Wj j is a coupling factor representing the strength of the interaction between molecules i and j, corresponding to a weighting of the effect of each molecule j on molecule i vis-à-vis the result of the set of combined effects of all molecules j exerting an eff and on i.
- the relationship between the variables Xj and X j , two by two is established by a sigmoid relationship comprising a delay factor associated with a linear decay function.
- the present invention also relates to a method for determining the mode of action of a xenobiotic, consisting in implementing a method for analyzing a network of molecular interactions in a biological system, such as those described above, under the following conditions: (i) the biological system in which a network of molecular interactions is studied is affected by the action of xenobiotic; (ii) the state to be modified "chosen in step C), corresponds to a state observed experimentally before the administration of said xenobiotic; (iii) the modifications to be made during step D) are identified for that the calculation carried out in step D) b) shows an evolution of the system towards a state close to the state observed after administration of the xenobiotic.
- Another aspect of the invention is a method of predicting undesirable effects of treatments applying a dynamic model of a network of molecular interactions in a biological system, by implementing a computer system.
- steps A to E we analyze on parts of the graph representative of known physiological functions, by simulations using the same method as in the previous aspects of the invention (steps A to E, possibly A to Ek when steps D and E are repeated iteratively by applying the stimuli on combinations of vertices up to k vertices), the consequences of the application of the treatment on these target molecules.
- This analysis consists in identifying the possible evolutions of these sub-parts of graphs towards new states close to other reference pathological states (as defined by the experimental observation of these pathological conditions, according to methods similar to what is described. upper).
- observation during simulations of the evolution of the apoptosis subgraph towards a final state having close proximity to a reference state of this graph corresponding to an activation of this physiological pathway makes it possible to predict an effect of cellular toxicity of the treatment in the tissue or tissues concerned.
- This aspect of the invention therefore consists in implementing an analysis method as described above, under the following conditions: (i) the biological system in which a network of molecular interactions is studied is concerned by the treatment; (ii) the modifications of step D) a) correspond to the modifications of the levels or activity levels of the target molecules observed or desired during the application of the treatment; (iii) step D) b) of calculating the evolution of the biological system is followed by an analysis of sub-parts of the system corresponding to known physiological functions, in order to identify any evolutions of these sub-parts towards states close to reference pathological states.
- the present invention also relates to a method for prioritizing potential therapeutic targets for a pathology, comprising identifying therapeutic targets by a method according to the invention, then predicting the possible undesirable effects of a treatment targeting these targets, and finally to determine the "therapeutic benefit / adverse effects" ratio of an action on each of the potential therapeutic targets.
- the present invention in its various aspects, is that it makes it possible to work on graphs or networks of interacting molecules comprising a large number of molecules.
- the number of variables Xi of the network of molecular interactions considered is therefore preferably greater than 100, greater than 200, or even greater than 300.
- the invention also relates to an analysis method as described above using the molecular interaction networks of the invention, said networks being associated to form a hypergraph of networks.
- the number of variables Xi of each network of molecular interactions is less than approximately 100 and the number of networks associated to form the hypergraph is between 2 and approximately 100.
- Another aspect of the invention is a method of extending the graphs from results of experimental screening for variations in the rate of expression or activity of molecules, applying a dynamic model of a network of molecular interactions in a biological system, by implementing a computer system.
- the steps and characteristics of the method are the same as previously, the only modification consisting in the following adaptation:
- the method is used to identify new molecular interactions. This can be achieved by coupling the method of the invention described above, with statistical methods for finding correlation between points in an n-dimensional space (for example factor analysis, hierarchical classifications, etc.) such as ( but not exclusively) those used to date to search for correlations of gene expression from the results of screening for messenger RNA on DNA chips ("clustering" of genes).
- cluster 3.0 developed by the Laboratory of DNA Information Analysis of Human Genome Center, http: //www.ims.u-tokyo.ac.ip/imswww/index- e.htmllnstitute of Médical Science, Universitv of Tokyo, au Japan (4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639 JAPAN).
- the "cluster 3.0" software is available on the website http: //bonsai.ims.u- tokyo.ac.jp/ ⁇ mdehoon/software/cluster/.
- the experimental data used can, for example, be that produced by messenger RNA expression screens on DNA chips. This coupling consists in using the parameters calculated by the implementation of the invention to recalculate a new matrix of experimental data for measuring the expression of the rate or activity of the molecules, by eliminating matrices of experimental results from originate the molecular interaction factors included in the parameterized dynamic model (such as the dynamic or inertial resistance component), then carry out correlation searches.
- This "cleaning" of the original result matrices consists in other words in eliminating the "statistical noise” linked to these factors, these factors then being considered as introducing distortions, in the measures actually observed of the expression rates. or activity of molecules, compared to what these measures would have been, from a theoretical point of view, in the absence of these factors.
- the dynamic resistance of the expression of a given gene A (therefore the inertia of the modification of the level of corresponding messenger RNA) to two distinct stimulations exerted by the molecules B and C (themselves distinct ) can vary, preventing before any "cleaning" of this type from highlighting both a correlation between the expression of A and the activity of the molecule B, and a correlation between the expression of A and the activity of molecule C.
- the invention therefore relates to the use of a dynamic model of a network of molecular interactions in a biological system capable of being obtained by a method as described above, to extend a graph. static whose vertices represent biological molecules and the arcs represent physico-chemical interactions between these molecules.
- FIG. 1 represents a synthetic diagram of the various steps of a method for identifying targets according to the present invention.
- the text within each rectangle corresponds to the abbreviation of the name of the protein as described in the text.
- the parameters of the relationships between the vertices of the graph and the simulations were calculated on the whole of this graph (example 4).
- Figure 3 shows graphical examples of calculated (triangles) and observed (squares) kinetics for some genes (example 4).
- Figure 3A ORF YBL015W (ACM).
- Figure 3B ORF YMR169C (ALD3).
- Figure 3C ORF YIL125W (KGD1).
- Figure 3D ORF YNL071W (PDA2).
- Figure 3E ORF YAL054C (ACS1).
- Figure 3F ORF YFL018C (LPD1).
- FIG. 4 represents a diagram of the graph constructed in example 5. This graph includes 133 molecules in the network of molecular interactions (133 vertices), and 407 molecular interactions between these molecules.
- Figure 5 shows examples of parameterization curves, in which the kinetics measured experimentally are represented in white and the kinetics calculated by simulation are represented in black, for some molecules (example 5).
- Figure 5A ICL 1 (YER065C).
- Figure 5B IDH1 (YNL037C).
- Figure 5C ACM (YBL015W).
- Figure 5D PCK1 (YKR097W).
- FIG. 6 represents a classification scheme of the molecules of the network by hierarchical classification of the distances calculated between on the one hand the state to be reached and on the other hand the states obtained by simulation
- the ordinates correspond to the calculated distance values.
- the 133 molecules of the network are classified from left to right from that associated with the shortest distance to that associated with the highest distance, each point corresponding to a molecule in the network.
- i be a given molecule of the network, and Xj its quantity (or its concentration) within the biological system studied.
- x i0 be the experimental measurement actually carried out from i to a "standard state" of the system biological, used during measurements.
- XH be the experimental measure actually carried out from i at an instant t. The variable used is:
- the standard efaf is a measurable condition used to perform biological measurements, against which all other measurements are quantified. It can correspond to an artificial state of the system, for example to a pool of several biological samples taken under different experimental conditions (artificial state), or to a really observable state (not artificial) of the system. This variable corresponds well to the type of biological measurements actually achievable.
- the measurement actually carried out for each RNA at a given experimental time t is the ratio of the signal emitted by the hybridization from RNA present in the biological sample at time t on the signal emitted by RNAs of the same type present in the sample at a standard state of the biological system studied (for example the initial time of the biological experiment).
- RNA molecules not being directly measurable because it depends on experimental parameters not directly controlled (yield of the labeling reactions of the probes, yield of hybridizations on the chips, etc. , these parameters differing in a non-predictable way between two RNAs of different types given).
- the quantity of signal measured in the standard state therefore serves as a measurement standard for that at other times, based on the assumption that for a given type of RNA, the experimental parameters influencing the signal finally emitted are the same.
- Xi therefore corresponds directly to the quantitative biological measures that are really productive in the current state of molecular biology techniques.
- Variables Xj, Xj etc. are therefore equal to (xit / xio), (jt / jo) etc. - (2)
- n be vertices ji, j 2 , ..., jn of the graph (corresponding to n molecules of the network) which act on a vertex i (orientation of the graph from j to i).
- Inertial term of these relations corresponds to a continuous function of the Xj. This term has an inertial component.
- inertia we mean the fact that Xj presents a resistance to change following a variation of Xj: more precisely, this term of the relation must make it possible to account for the behavior according to variables: following a given variation of one or more of X j , the speed of variation of Xi will be initially low, then accelerate gradually.
- the relationship between Xj and X j is established by an inertial relationship arising from that of the harmonic oscillator in physics associated with a damping factor large enough to limit the oscillation to a single cycle .
- Xi variable associated with the molecule i dX
- / dt derivative of Xj as a function of time d 2
- Xj / dt 2 second derivative of Xj as a function of time
- Xj variable associated with the molecule j
- mj inertia of i (resistance to change)
- ⁇ y damping (governs the return to the equilibrium state of Xj).
- ⁇ pulsation (response time of Xj to the stimulus represented by X j )
- wy coupling factor representing the strength of the interaction between molecules i and j, corresponding to a weighting of the effect of each molecule j on the molecule i with respect to the resultant of the set of combined effects of all the molecules j having an effect on i.
- Xi; dXj / dt; d 2 Xj / dt 2 and Xj are data supplied experimentally or directly calculated from these data. For example, this data can be obtained by screening for mRNA expressions.
- the relationship between Xj and the set of Xj is defined by a sum whose weighting includes a variable term over time and explicitly taking into account the respective velocities of the modifications of Xj, velocities represented by the derivatives: dX j (t) / dt, denoted in the sequence dX j / dt.
- dX j (t) / dt velocities represented by the derivatives: dX j (t) / dt, denoted in the sequence dX j / dt.
- ay is directly calculable, from experimental quantitative data, for each experimental time. This weighting factor varies over time.
- the relationship between Xj and the set of X j is defined by a sum whose weighting includes a variable term over time and which takes explicit account of the respective accelerations of the modifications of the Xj, accelerations represented by the second derivatives: d 2 X j (t) / dt 2 denoted in the following d 2 X j (t) / dt 2 .
- d 2 X j (t) / dt 2 denoted in the following d 2 X j (t) / dt 2 .
- the relationship between Xi and the Xj is established by a sigmoid relationship comprising a delay factor associated with a linear decay relationship.
- the relation associated with the edges corresponding to an inhibitory molecular interaction could be the opposite of that associated with the edges corresponding to an activating interaction (sigmoid curve decreasing or increasing, respectively), this characteristic being directly obtained during the calculation of the parameters. , by their positive or negative signs.
- K ⁇ . [1 / (1 + e- ⁇ w0J - bl )] corresponds to the inertial term
- the term : corresponds to the term of return to the initial state.
- Xi variable associated with vertex i.
- Xj variable associated with vertex j.
- wij weighting factor of the effect of j on i when several vertices (ji. J 2 , • Jn) act on the vertex i.
- bi delay factor
- the weighted result of the combination of the effects of the set of Xj on Xj is included from the start.
- the combined effect of all X j is also represented here by a weighted sum, in the term inertial.
- the relationship between the Xj and the X j is established if ilary to the previous aspect, but with the introduction of an additional weighting factor ay as defined by equations (4) or (7) above.
- the parameter ay is a variable term over time and takes explicitly into account the respective velocities of the modifications of X j , or the respective accelerations of Xj, depending on whether a is defined by equation (4) or equation (7) , respectively. Since the times at which the experimental measurements are made are long compared to the accelerations, it is preferable define the weighting factors ay with respect to the velocities (equation
- Equation (4)
- Equation (7)
- E ⁇ I [Xii (t) -X 2i (t)] 2 .dt it with: Xij (t): value of Xj at time t calculated by simulation with: X 2 j (t): value of X; at time t measured experimentally.
- V square root of the term in square brackets: []. X ⁇ (t) and X 2 i (t) being defined as above.
- the graph (or network) is entirely determined by the mathematical relations associated with the edges of the graph: it is a fully deterministic network.
- the corresponding graph is oriented.
- the network can be represented explicitly by the implementation of techniques for representing neural networks used in artificial intelligence. It is a non-Boolean network, neither Bayesian, nor organized in layers, allowing to represent redundancies of circuits and feedback loops. This deterministic network allows the implementation of simulations without significant computation cost, even for a very large graph.
- propagation methods include the Neural Network Toolbox 4.0.2 software, developed in the computing environment
- Propagation is inherent in the learning methods mentioned in Example 2.
- the simulation step therefore consists in using the propagation method implemented in step B, or any other propagation method deemed adequate by humans. art.
- thresholding procedures are associated with the propagation method chosen, in order to reduce the divergences (therefore to improve the convergences, that is to say the reliability). These can relate to:
- a lower threshold i.e. any value of a predicted variable below 0 is reduced to 0 (or to a minimum background noise value if this can be defined by experimental data):
- a higher threshold by imposing a maximum threshold on the values of Xj which can be obtained during simulations (for example by fixing a maximum threshold corresponding to a multiplicative factor (which can not exclusively be fixed between 1 and 10) of the maximum values observed experimentally for each Xj; this factor can possibly be defined during simulations of real experimental results available by testing several values of this factor -
- the introduction of constraints in the loops during simulations This can be achieved by several methods, not exclusive some of others, this list is not exhaustive.
- these simulations aim to analyze the biological system as a whole, and therefore to respond to the issues mentioned above.
- These simulations therefore consist in describing the evolution of all the molecules in the network of molecular interactions over time following the initial "virtual” stimuli, including, if it turns out to be biologically important, until a new state balance of the graph.
- These “virtual” stimuli can be applied systematically to each vertex of the graph, but it is also possible to carry out several stimuli at the same time, or sequentially, on several vertices.
- Proximity of the states of the graph A statistical calculation of proximity between each final state calculated and the desired state, or between each final state and the state to be modified, is carried out. It allows, for each vertex, to associate a statistical criterion (proximity obtained) with the vertex on which the stimulus was exerted and with the stimulus exerted on this vertex.
- the distance between two states of a graph can be calculated as follows:
- Example 4 modeling and simulations from a static graph of 116 molecules Biological data used:
- a static graph corresponding to a network of molecular interactions in yeast was constructed by manual entry in a flat file txt type (without implementing a system automated), using data from the KEGG database: Kyoto Encyclopedia of Genes and Genomes, free access data at the internet address http.7 / www. ⁇ enome.ad.ip / keqq / keqq2.html.
- This graph is more particularly centered on the mechanisms of cellular respiration (glycolysis, neoglucogenesis, metabolism of Pyruvate and acetyl CoA, etc.). It includes 116 molecules, enzymes or transcription factors. It includes 329 uni- or bi-directional interactions between these molecules.
- each rectangle represents a protein.
- the letters in the rectangle are the usual abbreviations of the name of the protein, according to the nomenclature of KEGG and SGD: Saccharomyces Genome Database developed by the Department of Genetics at the School of Medicine, Stanford University, USA.
- This table presents the static graph data in a form which can be directly used by a computer modeling and simulation system as described in the invention.
- the graph represents the 329 interactions between the 116 molecules in the network. Interactions are represented between the molecules two by two.
- This table also establishes the correspondence between the ORF (open reading fram ⁇ ) codes of the SGD database (Dolinski, K., Balakrishnan, R., Christie, KR, Costanzo, M. C, Dwight, SS, Engel, SR, Fisk, DG, Hirschman, JE, Hong, EL, Issel-Ta ⁇ er, L., Sethuraman, A., Theesfeld,
- ORF codes are unique for a given protein and allow it to be identified without any ambiguity. They also make it possible to establish an unambiguous link with the results of screening on DNA chips (correspondence of the nucleic sequences of the corresponding messenger RNAs).
- This publication describes an experiment in the culture of yeasts under conditions where the concentration of glucose in the culture medium gradually decreases (due to its use by yeasts for fermentation, since glucose is not added to the culture at any time. experience). Over time, the yeasts change their metabolism, their respiratory system going from functioning in fermentation to functioning in aerobic respiration.
- This yeast culture has been studied over time, in particular by the practice of expression screening of almost all of the yeast messenger RNAs on DNA chips. These screenings were carried out at successive times, the results therefore producing an expression level kinetics for each messenger RNA.
- the results show variations in the level of expression of a certain number of messenger RNAs over time, these being more particularly numerous among the messenger RNAs of cellular respiration proteins, a large part of which is represented in the graph described above. Under these experimental conditions, the graph that we have constructed therefore presents a dynamic evolution over time, which is represented by the kinetics of the molecules of the network (therefore the vertices of the graph).
- each line corresponds to a molecule of the graph
- the first column identifying the ORF (open reading frame) by its SGD code
- Tables 6 to 8 below give examples of data for some molecules of the graph, data extracted from the page http://cmgm.stanford.edu/pbrown/explore/array.txt.
- Tables 6 to 8 Examples of DNA microarray screening data for 2 of the molecules of the graph, from the results of the article: DeRisi JL, lyer VR, Brown PO: Exploring the metabolic and genetic control of gene expression on a genomic scale, Science. 1997 Oct 24; 278 (5338): 680-6. Full data are available on the web page: http://cmgm.stanford.edu/pbrown/explore/array.txt. Given the size of the complete data table, only part of it is shown here. Those skilled in the art will very easily be able to recover the data corresponding to the other molecules of the graph used in this example, on this internet page which corresponds directly to the table of all the data.
- the expression screens of the messenger RNAs were carried out every two hours, for 12 hours, which corresponds to 7 experimental times (the initial time plus the 6 following times). These correspond to notations 1 to 7. The reader will find all the explanations corresponding to obtaining these measurements in the article cited in reference.
- Name abbreviation of the name of the gene (according to SGD)
- ORF open reading frame code (according to SGD)
- G experimental condition corresponding to the "standard state" of the graph as described in the invention. This standard state, in this series of experiments, corresponds to the initial state of yeast culture. G1, G2, G3, G4, G5, G6,
- G7 all correspond to the same standard biological sample.
- R states of the graph at various experimental times.
- R1 corresponds to the initial state of culture of the yeasts (same biological sample as G1), at time TO.
- R2 TO + 2.5 hours
- R3 TO + 4 hours
- R4 TO + 6 hours
- R5 TO + 7.5 hours
- R6 TO + 9.5 hours
- R7 TO +
- G2-Bkg measurement of G2 minus the background noise during experimental measurements.
- G3-Bkg measurement of G3 minus the background noise during experimental measurements.
- G4-Bkg measure of G4 minus the background noise during experimental measurements.
- G5-Bkg measurement of G5 minus the background noise during experimental measurements.
- G6-Bkg measurement of G6 minus the background noise during experimental measurements.
- G7-Bkg measurement of G7 minus the background noise during experimental measurements.
- the variations in the measured values are linked to the variations in the yields of the various reactions used in the measurement method (DNA chips) and justify the use of a standard state as a measurement reference.
- R1-Bkg measurement of R1 minus the background noise during the experimental measurements.
- R2-Bkg measurement of R2 minus the background noise during experimental measurements.
- R3-Bkg measurement of R3 minus the background noise during the experimental measurements.
- R4-Bkg measurement of R4 minus the background noise during the experimental measurements.
- R5-Bkg measurement of R5 minus the background noise during the experimental measurements.
- R6-Bkg measurement of R6 minus the background noise during experimental measurements.
- R7-Bkg measurement of R7 minus the background noise (experimental noise).
- Steps A and B were implemented as follows:
- the parameters were calculated from the experimental data by a classic method of learning neural networks, more precisely from the algorithms of the Runge Kutta method of back propagation through time (BPTT: back propagation through time ). The calculations were performed in double precision.
- the effectiveness of the method was checked. Indeed, the average divergence result (global relative error) of the simulations compared to the experimental data is approximately 0.30, this divergence being essentially concentrated on 8 vertices (molecules) of the graph, for this series of data, whereas for the remaining 108 vertices (molecules), the divergence is very small.
- This overall relative error result shows that the kinetics calculated during the simulations are close to the real data, since random kinetics would have given an error calculation greater than 1.
- V square root of the term in square brackets [].
- X-ii (t) value of X calculated at time t of the simulation
- X 2 i (t) value of X measured experimentally at time t.
- ⁇ sum of values at different times t
- the rate of non-reproducibility of the experimental data can be estimated by the ratio R1. Ratio of experimental data (Table 8), and is overall 14% in this example.
- table 9 gives the details of the calculations of the divergences of the set of kinetics during the simulations for all of the molecules of the network, in the form of a summary table.
- Xii (t) value of Xi calculated at time t of the simulation
- X 2 i (t) value of Xi measured experimentally at time t
- This calculation amounts to calculating the difference in integral between the curves of the kinetics observed experimentally and the kinetics calculated during the simulations. It therefore concerns both the whole kinetics and the final state.
- this variant does not of course consist in removing from the graph the "input" vertices, but in imposing that their kinetics remain the kinetics measured experimentally, this only during the simulations practiced during the error calculations of the procedure for calculating parameters by back propagation over time.
- the parameters of the mathematical relations connecting these vertices to other vertices of the graph are therefore calculated, as for all the other edges of the graph, and the dynamic model finally obtained includes these vertices.
- FIG. 3 gives by way of example the kinetics measured experimentally and the kinetics calculated by simulation for a few genes representative of the set of results obtained by the implementation of this variant.
- Example 5 Modeling, simulations and validation of predictive capacity from a static graph of 133 molecules
- This example shows the implementation of the whole method (steps A and B or A ', then C, D, E) and its predictive effectiveness in an application similar to the identification / selection of therapeutic targets.
- a static graph corresponding to a network of molecular interactions in yeast was built according to the same principles as in Example 4. This graph more particularly includes the graph of Example 4, but with molecules and interactions additional. It includes 133 molecules, enzymes or transcription factors. It includes 407 uni- or bi-directional interactions between these molecules.
- Glucose its concentrations over time were measured by the authors of the publication, at the same time as those for which the messenger RNA expression measurements were taken. The corresponding concentrations are given graphically in FIG. 4 of the cited article. In order to express these values as a ratio, each value of the concentration of Glucose in the culture medium at a given time was divided by the concentration of Glucose at the initial time of the experiment (this in order to measure the ratios by compared to the same frame of reference as for messenger RNA measurements, the ratios of which are expressed by the ratio of the measurement at time t divided by the measurement at the initial time).
- Acetate and Glutamate the concentrations of these molecules were not measured by the authors. It was therefore decided to extrapolate values for these molecules from knowledge of the biological system studied and from the description of the experiments in the article. Insofar as this experiment is essentially based on the progressive fall in the concentration of glucose in the culture medium and since the other parameters of the culture medium are at first approximation considered to be constant, it was considered that the concentrations of Glutamate and of Acetate, respectively, were constant during the experiment.
- Table 12 Values of the variables associated respectively with the molecule of the graph: Glutamate and with the molecule of the graph: Acetate
- Steps A and B or step A ' were implemented in a similar manner to Example 4:
- the initial state which corresponds to the experimental measurements at time TO, would be a stable state if the culture medium had been kept constant by supplementing it with glucose.
- we did not define a final state of the graph which would correspond to a return to the initial state following a glucose supplementation after time 7 (TO + 12 hours).
- the only experimental data to have been used for the calculation of the parameters were therefore the data actually described in the article and present on the website of Stanford University at the address: http://cmqm.stanford.edu /pbrown/explore/array.txt, and corresponding to the molecules of the network, without any extrapolation other than that concerning the molecules Glucose, Glutamate and Acetate described above.
- the learning step for calculating the parameters was fixed at 16 minutes.
- Step C was implemented as follows:
- the messenger RNA expression screening data corresponding to this state were therefore those of the initial time used for the construction of the
- Stage D was implemented as follows
- This step consists in practicing iterative simulations, as described in the invention.
- the yeast strain with the deletion of the TUP1 gene initially differs from the “wild” strain only in this deletion. This deletion returning to a constant inhibition of the expression of the TUP1 gene, the simulations consisted in simulating, in an iterative manner, the constant inhibition of each of the 133 molecules of the network, and in performing a calculation of propagation over the time of this inhibition within the network. For each simulation, a single molecule in the network was inhibited, since the state to be reached corresponds to an evolution of the modeled biological system (the yeast strain) following a single inhibition (deletion of the TUP1 gene). We therefore carried out 133 simulations.
- the article's experimental data and the messenger RNA expression screening data concerning the strain exhibiting the TUP1 gene deletion (available on the Stanford University website at 1 'address: http://cmgm.stanford.edu/pbrown/explore/tupsearch.html).
- the deletion of the gene was incomplete in this biological experiment, the ratio: [level of expression of the TUP1 gene in the deleted strain] / [level of expression of the TUP1 gene in the wild strain] being equal to 0.1 to a measurement , and 0.45 when replicating the measurement (average: 0.28). In the case of a complete deletion, this ratio would have been theoretically equal to 0, and equal in practice to the experimental measurement noise.
- the inhibition was therefore defined numerically, for each molecule of the graph, as the multiplication by a factor of 0.1 of the level of expression of this molecule at the initial time (state to be modified as defined above), this factor corresponding to the value of the strongest inhibition measured experimentally for this gene.
- the molecule i was different: the effect of each inhibition by a factor of 0.1 on each molecule in the graph was tested systematically.
- the inhibition was imposed as a constant over time during the simulations: thus, during the propagation calculations, a possible propagation feedback on the inhibited molecule i
- step D was implemented as described in the invention, without any notable particularity, and without presenting any particular difficulty for those skilled in the art.
- the simulation calculations, consisting of propagating the initial inhibition over time, were performed using the same principles and the same tools as the simulations that are part of the parameter calculation procedure.
- step D Each of the 133 simulations of step D thus resulted, at time 12 hours (duration of the simulated propagation), in the calculation of a new numerical value associated with each molecule of the network, defining a state of the graph: "state obtained by simulation ". We therefore obtained 133 different “states obtained by simulation”.
- Step E was implemented as follows:
- This step consists in prioritizing the molecules of the graph, and the effects exerted on these molecules during simulations, with reference to the greater or lesser proximity of the resultant of these effects with a state of the graph to be reached.
- the state to be attained was the state of the yeast strain having a deletion of the TUP1 gene described above.
- the value of Xj corresponding to Glucose for the strain exhibiting the deletion of the TUP1 gene was fixed at 1 since the screening was carried out on a culture in the presence of glucose.
- the Xj values corresponding to Glutamate and Acetate for the strain exhibiting the deletion of the TUP1 gene were set to 1 since the screening was carried out on a strain in a culture medium identical to that of the wild strain at time 0 (between the two cultures, the metabolite ratios in the culture medium are therefore equal to 1).
- Table 14 Complete list of the values of the state to reach as defined above
- Step E then consists in calculating the distance between on the one hand “the state to reach” of the graph, and on the other hand each of the 133 “states obtained by simulation” of the graph obtained in step D.
- This distance calculation is described above (proximity of the states of the graph) and does not pose any particular difficulty for those skilled in the art. It consists in comparing two states of the graph by comparing two by two the set of values Xi associated with each molecule i of the graph.
- Stage 1 consisted in carrying out a first classification by distance calculations in a conventional manner:
- Order 1 distance sum of the absolute values of the differences between the values of Xj measured experimentally during the deletion of the TUP1 gene (X, 2 in the formula below) and the values of Xj measured by simulation (X ⁇ in the formula below) i
- Stage 2 consisted, following this first classification, in refining it.
- This classification step is given here by way of example but is not essential for the implementation of step E.
- the sensitivity was then defined, for each of the 133 simulations, as the number of molecules of group Bj actually present in group A. This amounts to evaluating, for the quantitatively significant variations in expression of the molecules (greater than a factor of 2) if the simulation actually induces the variations present in the experimental data of the state to be reached. The higher the value of the sensitivity, the smaller the distance between the two states of the compared graph.
- the specificity was then defined, for each of the 133 simulations, as the number of molecules of group Bi absent from group A. This amounts to assessing, for the quantitatively significant variations in expression of the molecules (greater than a factor of 2) if the simulation does not induce variations of expression absent in the experimental data of the state to be reached. The lower the value of the specificity, the smaller the distance between the two states of the compared graph.
- the difference sensitivity - specificity therefore amounts to evaluating the distance on the combined criteria of induction by the simulation of the variations of expression present in the state to be reached and non-induction by the simulation of variations of expression absent in the state to reach.
- sensitivity - specificity is also very simple and can for example be calculated manually, or automatic by an Excel software table (Microsoft) or any other equivalent software.
- the relative overall error-learning calculation was 25.90%, which is satisfactory.
- This global relative error result shows that the kinetics calculated during the simulations are close to the real data; random kinetics would have given an error calculation greater than 1.
- FIG. 5 gives by way of example the kinetics measured experimentally (in white) and the kinetics calculated by simulation (in black) for a few representative molecules of the set of results obtained by the implementation of this variant of steps A and B or from step A '.
- step B The calculation of the parameters carried out in step B therefore made it possible to obtain a dynamic graph which takes into account the experimental data used for the calculation.
- the result of the first classification step is summarized by way of example in the following table, in the form of a summary table.
- Each molecule in the network is classified by the distance between the target state of the graph and the state of the graph obtained by simulating constant inhibition by a factor 0.1 of this molecule.
- Table 16 Distances between the target state of the graph and the state of the graph obtained by simulating constant inhibition by a factor 0.1 of each molecule
- the classification in ascending order of the distances gives directly the classification of the molecules from the one whose inhibition is most likely to cause the state to reach of the graph to that whose inhibition is the less likely to cause it.
- the TUP1 molecule is classified in first position, which is the expected result. The method is therefore validated.
- FIG. 6 gives by way of example a schematic representation of this result of classification of the molecules of the network.
- Each point corresponds to a molecule in the network.
- the ordinates correspond to the calculated distance values.
- On the abscissa the 133 molecules of the network are classified from left to right from that associated with the lowest distance to that associated with the highest distance.
- the TUP1 molecule is classified in first position. This result is completely satisfactory.
- the experimental test of the first 5 molecules as classified here would therefore give a success rate of 100% for the selection of the relevant molecule.
- This classification also has the advantage of being able to define a “boundary” of all of the selected target molecules. Indeed, certain molecules of the graph do not send any interaction to another molecule (they are therefore “outputs” or “outputs” of the graph: “molecules outputs” in the rest of the text); the simulation of the inhibition of these molecules therefore does not cause the inhibition to propagate within the graph which therefore remains globally stable (since we considered that the initial state was stable). Therefore, these molecules are also classified in a contiguous group, from the 27th position to the 30th position. I_es molecules less well classified than the output molecules are therefore not of interest as a target molecule.
- the classification can be interpreted as follows: the best-placed molecule is the one whose simulation of inhibition results in the state of the graph closest to the state to be reached. For the following molecules, we gradually move away from the state to reach, by going towards the state to modify that we reach when we arrive at the output molecules (this state not being modified during the simulation of the inhibition of output molecules, except for the output molecule). Beyond the output molecules, the inhibition simulations lead to states of the graph which gradually move away from both the state to be reached and the state to be modified. We did end up with the selection of a limited number of target molecules (those that are better classified than the outputs, here 26 molecules), which are themselves prioritized in terms of priority. The classification of the TUP1 molecule shows that this ranking is satisfactory.
- Classification step 2 (the sensitivity - specificity calculation), although not essential given the above result, was then implemented in order to improve the classification of the targets. It was only applied to the molecules of the graph better classified than the output molecules in the previous step (26 molecules). The classification thus obtained is given in table 17.
- Table 17 Final hierarchy of the selected target molecules (the 26 molecules before the outputs) The molecules are classified in descending order of the arithmetic difference [sensitivity - specificity]. We see that TUP1 is the only molecule in the graph for which the sensitivity is greater than the specificity. This individualizes it from others on qualitative criteria. Here again, TUP1 is in first position in the hierarchy of potential targets.
- the TUP1 target molecule is ranked first. It is indeed the molecule of the graph whose inhibition by deletion of the gene induces the evolution of the biological system studied towards the state to be reached. The effectiveness of the method is therefore verified: selection of a limited number of target molecules, and relevance of the hierarchical classification of these targets.
- steps A and B, or A ', then C, D and E has been carried out more than ten times, and has systematically given similar results each time (with classification of TUP1 always in the first 5 molecules by the first classification mode). This shows without ambiguity the reproducibility of the method as well as its effectiveness.
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Physiology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biotechnology (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Theoretical Computer Science (AREA)
- Investigating Or Analysing Biological Materials (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Medicines That Contain Protein Lipid Enzymes And Other Medicines (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
FR0309557A FR2858446B1 (fr) | 2003-08-01 | 2003-08-01 | Methode d'analyse de reseaux d'interactions moleculaires biologiques |
PCT/FR2004/002064 WO2005013173A2 (fr) | 2003-08-01 | 2004-07-30 | Methode et systeme de selection de cibles therapeutiques par l'utilisation de reseaux dynamiques d'interactions moleculaires |
Publications (1)
Publication Number | Publication Date |
---|---|
EP1649405A2 true EP1649405A2 (fr) | 2006-04-26 |
Family
ID=34043752
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP04786022A Withdrawn EP1649405A2 (fr) | 2003-08-01 | 2004-07-30 | METHODE ET SYSTEME DE SELECTION DE CIBLES THERAPEUTIQUES PAR L'UTILISATION DE RESEAUX DYNAMIQUES D INTERACTIONS MOLECULA IRES |
Country Status (5)
Country | Link |
---|---|
US (1) | US20060235670A1 (fr) |
EP (1) | EP1649405A2 (fr) |
CA (1) | CA2534401A1 (fr) |
FR (1) | FR2858446B1 (fr) |
WO (1) | WO2005013173A2 (fr) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7667582B1 (en) * | 2004-10-14 | 2010-02-23 | Sun Microsystems, Inc. | Tool for creating charts |
CA2726175A1 (fr) | 2008-05-27 | 2009-12-23 | Memorial Sloan-Kettering Cancer Center | Modeles de perturbations combinatoires de systemes biologiques vivants |
EP2715586A2 (fr) * | 2011-06-01 | 2014-04-09 | Albert-Ludwigs-Universität Freiburg | Procédé de modélisation, d'optimisation, de paramétrage, de test et / ou de validation d'un réseau dynamique ou de perturbations de réseau |
EP2530615A1 (fr) * | 2011-06-01 | 2012-12-05 | Albert-Ludwigs-Universität Freiburg | Procédé de modélisation, optimisation, paramétrage, test et/ou validation des perturbations de réseau dynamique |
EP3298524A4 (fr) | 2015-05-22 | 2019-03-20 | CSTS Health Care Inc. | Mesures thermodynamiques portant sur des réseaux d'interaction protéine-protéine pour le traitement du cancer |
US10366324B2 (en) * | 2015-09-01 | 2019-07-30 | Google Llc | Neural network for processing graph data |
US11456053B1 (en) | 2017-07-13 | 2022-09-27 | X Development Llc | Biological modeling framework |
CN109635439B (zh) * | 2018-12-12 | 2023-05-05 | 新疆大学 | 基于传播动力学的扰动环境下生产网络连锁效应研究方法 |
CN116844645B (zh) * | 2023-08-31 | 2023-11-17 | 云南师范大学 | 一种基于多视角分层超图的基因调控网络推断方法 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2001286880A1 (en) * | 2000-11-28 | 2002-06-11 | Physiome Sciences, Inc. | System for modeling biological pathways |
AU2002355996A1 (en) * | 2001-08-16 | 2003-03-03 | Biotech Research Venture Pte Limited | Method for modelling biochemical pathways |
-
2003
- 2003-08-01 FR FR0309557A patent/FR2858446B1/fr not_active Expired - Fee Related
-
2004
- 2004-07-30 EP EP04786022A patent/EP1649405A2/fr not_active Withdrawn
- 2004-07-30 WO PCT/FR2004/002064 patent/WO2005013173A2/fr active Application Filing
- 2004-07-30 CA CA002534401A patent/CA2534401A1/fr not_active Abandoned
-
2006
- 2006-01-31 US US11/342,707 patent/US20060235670A1/en not_active Abandoned
Non-Patent Citations (1)
Title |
---|
See references of WO2005013173A2 * |
Also Published As
Publication number | Publication date |
---|---|
WO2005013173A2 (fr) | 2005-02-10 |
CA2534401A1 (fr) | 2005-02-10 |
WO2005013173A3 (fr) | 2005-09-29 |
FR2858446B1 (fr) | 2007-11-09 |
FR2858446A1 (fr) | 2005-02-04 |
US20060235670A1 (en) | 2006-10-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Casalino et al. | AI-driven multiscale simulations illuminate mechanisms of SARS-CoV-2 spike dynamics | |
Ragoza et al. | Protein–ligand scoring with convolutional neural networks | |
Hill et al. | Inferring causal molecular networks: empirical assessment through a community-based effort | |
JP6356359B2 (ja) | アンサンブルに基づいたリサーチ・レコメンデーションシステムおよび方法 | |
Mathur et al. | Gene set analysis methods: a systematic comparison | |
Yang et al. | Openrec: A modular framework for extensible and adaptable recommendation algorithms | |
Nicolaou et al. | De novo drug design using multiobjective evolutionary graphs | |
Ma et al. | CGI: a new approach for prioritizing genes by combining gene expression and protein–protein interaction data | |
Dawson et al. | Coarse-grained modeling of RNA 3D structure | |
Rudolph et al. | Elucidation of signaling pathways from large-scale phosphoproteomic data using protein interaction networks | |
Jaeger et al. | Mixed effect models for genetic and areal dependencies in linguistic typology | |
US8571803B2 (en) | Systems and methods for modeling and analyzing networks | |
Meyer et al. | Advances in systems biology modeling: 10 years of crowdsourcing DREAM challenges | |
US20170329904A1 (en) | Database and data processing system for use with a network-based personal genetics services platform | |
Hooft et al. | Discovering collective variables of molecular transitions via genetic algorithms and neural networks | |
US20110119221A1 (en) | Method, system and software arrangement for reconstructing formal descriptive models of processes from functional/modal data using suitable ontology | |
US20060235670A1 (en) | Method and system for selecting therapeutic targets using molecular interaction dynamic networks | |
Zhou et al. | How do tumor cytogenetics inform cancer treatments? dynamic risk stratification and precision medicine using multi-armed bandits | |
McComb et al. | Generalized pharmacometric modeling, a novel paradigm for integrating machine learning algorithms: a case study of metabolomic biomarkers | |
Abe et al. | UNMF: a unified nonnegative matrix factorization for multi-dimensional omics data | |
EP1934873A1 (fr) | Procede pour determiner l'etat d'un ensemble de cellules et systeme pour la mise en oeuvre dudit procede | |
US20240303544A1 (en) | Graph database techniques for machine learning | |
Oloulade et al. | Cancer drug response prediction with surrogate modeling-based graph neural architecture search | |
CN114171110A (zh) | 一种基于联合似然的孟德尔随机化分析方法 | |
Caulk et al. | Robust latent-variable interpretation of in vivo regression models by nested resampling |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
17P | Request for examination filed |
Effective date: 20060126 |
|
AK | Designated contracting states |
Kind code of ref document: A2 Designated state(s): AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LI LU MC NL PL PT RO SE SI SK TR |
|
AX | Request for extension of the european patent |
Extension state: AL HR LT LV MK |
|
DAX | Request for extension of the european patent (deleted) | ||
17Q | First examination report despatched |
Effective date: 20111117 |
|
RAP1 | Party data changed (applicant data changed or rights of an application transferred) |
Owner name: HELIOS BIOSCIENCES |
|
PUAJ | Public notification under rule 129 epc |
Free format text: ORIGINAL CODE: 0009425 |
|
32PN | Public notification |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 2524 DATED 24-03-2014) |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN |
|
18D | Application deemed to be withdrawn |
Effective date: 20140201 |