WO2021185936A1

WO2021185936A1 - System and method for generating personalized treatments

Info

Publication number: WO2021185936A1
Application number: PCT/EP2021/056883
Authority: WO
Inventors: Chen Zhao; Juliane Winkelmann; Barbara SCHORMAIR; Konrad OEXLE; Axel STELLBRINK
Original assignee: Chen Zhao; Juliane Winkelmann; Schormair Barbara; Oexle Konrad; Stellbrink Axel
Priority date: 2020-03-17
Filing date: 2021-03-17
Publication date: 2021-09-23

Abstract

The present invention is directed to a system and method for an automated generation of a treatment, such as a composing of a customized composition, such as a pharmaceutically active composition. It can comprise steps of inputting genetic data of a plurality of individuals; and inputting time stamped data of the plurality of individuals; combining the genetic data and the time stamped data of each individual to combined data; establishing and/or training a multi-layer neural network on the basis of the combined data; and predicting a disease risks and/or treatment response of the customized composition.

Description

System and method for generating personalized treatments

Field

The present invention relates to a system and a method for an automated and customized generating or predicting of a personalized treatment, such as composing a pharmaceutically active composition, particularly for genetic diseases.

Introduction

It has been tried to help patients by data processing approaches that take into consideration patient data and further health data. In general, genetic data of a person can be of particular interest in order to further specify a disease and to individualize treatment, such as by applying customized pharmaceuticals. However, such information is usually too unspecific as environmental and epigenetic conditions and a large number of time-dependent individual factors aggregate for the onset or further progress of a disease.

In EP 2554679 A1 there is described a cardiovascular disease risk assessment. This is based on the linear additive model of genetic effects. Moreover, it applies health data only as time-independent scores.

In Esteban et al., Predicting Clinical Events by Combining Static and Dynamic Information using Recurrent Neural Networks (arXiv: 1602.02685v2 [cs.LG] 17 Nov 2016), static clinical data (e.g. patient gender, blood type, etc.) was combined with sequences of data that were recorded during multiple hospital visits (e.g. medications prescribed, tests performed, etc.). Recurrent Neural Networks (RNNs) had proven to be very successful for modelling sequences of data in many areas of Machine Learning. In that work an approach was presented based on RNNs, specifically designed for the clinical domain, that combines static and dynamic information in order to predict future events. In a database collected in the Charite hospital in Berlin complete information concerning patients that underwent a kidney transplantation were collected. After the transplantation three main endpoints could occur: rejection of the kidney, loss of the kidney and death of the patient. It was the goal to predict, based on information recorded in the electronic health record of each patient, whether any of those endpoints will occur within the next six or twelve months after each visit to the clinic. Different types of RNNs that had been developed for that work were compared with each other, with a model based on a feedforward neural network and with a logistic regression model. It was found that the RNN that had been developed based on gated recurrent units performed best. The same models were tested in a second task, i.e., next event prediction, and it was found that the model based on a feedforward neural network now outperformed the other models. It was concluded that long-term dependencies are not relevant in the task.

A recurrent neural network for the prediction time-depend risks was reported by Choi et al. who used it in recurrent neural network models for early detection of heart failure onset (J Am Med Inform Assoc. 2017; 24: 361-370). In this approach, time-to-event risks were modelled as binary outcomes in few given temporal windows and crossentropy was used as cost function for training the network.

Kumardeep et al., Deep Learning based multi-omics integration robustly predicts survival in liver cancer (Clin Cancer Res. 2018; 24: 1248-1259) have tried to combine a genuine time-to-event cost function (Cox-PH) with a neural network. However, the cost function was not really part of their neural network. They used their network as an autoencoder to integrate different multi-omics data into new features, and only then used Cox-PH regression to select from the new features. For the actual risk prediction, they did not use a network but performed k-means partitioning and SVM analysis.

Khera et al., Polygenic Prediction of Weight and Obesity Trajectories from Birth to Adulthood. Cell. 2019;177:587-596. e9) derived a genome-wide polygenic score (GPS) for predicting weight and obesity (trajectories) from genome-wide genotype data. To do so they used an algorithm (LDPred, see Modeling Linkage Disequilibrium Increases Accuracy of Polygenic Risk Scores, Am J Hum Genet. 97: 576-592) that is based on the linear additive model of genetic effects but does not consider non-linear relationships between these effects. They did not use a multi-layered neural network. Concerning trajectories of weight development. They only showed that the top and bottom GPS deciles remained associated with over- and underweight, respectively. They did not include any time-to-event prediction.

Besides the before discussed Kumardeep et al., Deep Learning based multi-omics integration robustly predicts survival in liver cancer (Clin Cancer Res. 2018; 24: 1248- 1259), DeepSurv (DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network. BMC Med Res Methodol. 2018; 18: 24) has some similarity. DeepSurv implements a deep neural network with Cox-PH cost function. DeepSurv shares shortcomings with the autoencoder method which are overcome by our invention. DeepSurv is a sequential network that does not consider time-series input data, its Cox-PH cost function cannot adequately include time-varying effects, and it does not predict time-dependent risks.

In Kvamme et al., Time-to-Event Prediction with Neural Networks and Cox Regression (Journal of Machine Learning Research 20 (2019) 1-30), new methods for time-to-event prediction were proposed by extending the Cox proportional hazards model with neural networks. Building on methodology from nested case-control studies, a loss function that scales well to large data sets and enables fitting of both proportional and non proportional extensions of the Cox model was proposed. Through simulation studies, the proposed loss function is verified to be a good approximation for the Cox partial log- likelihood. The proposed methods were compared to existing methods on real-world data sets and was found to be highly competitive, typically yielding the best performance in terms of Brier score and binomial log-likelihood.

Ren et al., Deep Recurrent Survival Analysis (arXiv: 1809.02403v2 [cs.LG] 13 Nov 2018), is directed to a survival analysis for modeling time-to-event information with data censorship handling, which was widely used in many applications such as clinical research, information systems and other fields with survivorship bias. Many tools and works had been proposed for survival analysis ranging from traditional statistic methods to machine learning models. However, the methods either utilized counting-based statistics on the segmented data, or had a pre-assumption on the event probability distribution w.r.t. time. Moreover, few works considered sequential patterns within the feature space. In that paper, a deep recurrent survival analysis model is suggested which combines deep learning for conditional probability prediction at fine grained level of the data, and survival analysis for tackling the censorship. By capturing the time dependency through modeling, the conditional probability of the event for each sample allegedly predicts the likelihood of the true event occurrence and estimates the survival rate over time, i.e., the probability of the non-occurrence of the event, for the censored data. Meanwhile, without assuming any specific form of the event probability distribution, the model showed great advantages over the previous works on fitting various sophisticated data distributions. In the experiments on the three real world tasks from different fields, the model was said to significantly outperform the state-of-the-art solutions under various metrics.

Giunchiglia et al., RNN-SURV: a Deep Recurrent Model for Survival Analysis (ICANN 2018, pp 23-32), is directed to applied deep learning that enables medicine to become personalized to the patient at hand. A new recurrent neural network model called RNN- SURV was presented for personalized survival analysis. The model was considered to be able to exploit censored data to compute both the risk score and the survival function of each patient. At each time step, the network takes as input data the features characterizing the patient and an identifier of a time step, for creating an embedding, and outputting of a value of the survival function in that time step. Finally, the values of the survival function are linearly combined to compute the unique risk score.

In Zheng et al., SAFE: A Neural Survival Analysis Model for Fraud Early Detection (arXiv: 1809.04683v2 [cs.LG] 13 Nov 2018), online platforms are described that had deployed anti-fraud systems to detect and prevent fraudulent activities. However, there was considered a gap between the time that a user commits a fraudulent action and the time that the user is suspended by the platform. How to detect fraudsters in time was considered a challenging problem. Most of the existing approaches adopt classifiers to predict fraudsters given their activity sequences along time. As the main drawback of classification models was considered that the prediction results between consecutive timestamps are often inconsistent. Thus, a survival analysis was suggested in this paper based on a fraud early detection model, SAFE, which maps dynamic user activities to survival probabilities that are guaranteed to be monotonically decreasing along time. SAFE adopts a recurrent neural network (RNN) to handle user activity sequences and directly outputs hazard values at each timestamp, and then, survival probability derived from hazard values is deployed to achieve consistent predictions. Because the training data contained only the user suspended time instead of the fraudulent activity time the loss function of the regular survival model was revised to achieve early detection of fraud. Experimental results on two real world datasets allegedly demonstrated that SAFE outperformed both the survival analysis model and recurrent neural network model alone as well as state-of-the art fraud early detection approaches.

The before described processes are often cumbersome and time-consuming and result in an unsatisfying outcome.

Summary

In light of the above, it is an object of the present invention to overcome or at least alleviate the shortcomings of the prior art. More particularly, it is an object of the present invention to provide a method and a corresponding system for simulating and/or generating a customized and/or personalized treatment to a disease, such as a genetic disease. The present invention is directed to a system and a method for an automated simulation and/or generation of a treatment, such as a composing of a customized and/or personalized composition, such as a pharmaceutically active composition. The system and method can also be configured to predict a disease. It can comprise a genetic data base with genetic data of a plurality of individuals. The genetic data can comprise genomic data, genomic sub-data, data regarding relevant mutations, sequences, CNVs or SNPs data. The latter can be provided for each genetic disease separately in order to speed up computing. However, genomic data can also be used, applying a specific read out depending on the genetic disease at issue.

The term genetic disease is meant to comprise an health-impacting development of an individual that is trigged and/or influenced by the individual's genome or mutations of the genome or parts thereof.

A second data base can be arranged with time stamped data of the plurality of individuals. Time-stamped data in this context means information about incidents, health-relevant data, disease-relevant data, epigenetic data, phenotypic data, and somatic genetic data such as an aneuploidy, translocation, insertion, deletion, indel, and/or single nucleotide variation (SNV) etc. that is realized at a specific time and not already coded in the germline genome of an individual and any combination thereof.

The time of occurrence can be comprised in the time-stamped data or can be aggregated in time bins, as also mentioned below.

A third data base can be provided with treatment data comprising a plurality of treatments. The kinds of treatments are exemplified further below. The treatment data can comprise a plurality of applications for a plurality of customized pharmaceutically active compositions, one or more electro-magnetic treatment(s), one or more irradiation treatment(s) and any combinations thereof.

A machine learning component can be configured for training a predicting component for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment. The term predicting component is not intended to be interpreted in a limited manner but shall comprise a designation for a component, such as a routing or sub-routine, that is trained to serve later as a tool for the generation and/or simulation and/or prediction of an outcome of a treatment or multiple treatments or combined treatments. The predicting component can be further configured to generate and/or suggest a treatment on the basis of the machine learning as also exemplified further below. The machine learning component can be provided and/or configured for establishing and/or training a multi-layer neural network on the basis of the combined data. The data combination can be done in any appropriate way, such as by concatenation.

The neural network then also predicts a disease risks and/or treatment response of the customized composition. This prediction is useful for the composition of different components for treating the disease at a specific point in time or over a certain time frame.

The present invention can be a simulation system or method for simulating or predicting a treatment response of a treatment that is preferably automatically determined or generated.

The system can further comprise a re-adjustment component configured for a re adjustment of the customized treatment according to the prediction of the treatment response. The inventive method comprises a re-adjustment step accordingly.

The re-adjustment component can be configured for a modification of parameters of the treatment. Parameters can be the dosage parameters of compositions, the intensity or duration of electro-magnetic or irradiation treatments etc.

The re-adjustment can alternatively or additionally modify a kind of the treatment. As a kind of treatment, the nature of the treatment is meant, such as a chemical treatment by pharmaceutically active components, an irradiation and/or an electro-magnetic treatment.

Thus, the customized treatment can be a customized composition of a pharmaceutically active composition, an electro-magnetic treatment and/or an irradiation treatment.

It is also within the frame-work of the present invention to combine more than one (plurality) of components and then let the neural network predict the impact onto an individual. The impact can comprise prediction on the effect onto the individual and against the disease but also predictions about side-effects on each individual and any combination thereof. By this, the most effective or efficient composition can be composed.

The prediction can be output in a plurality of manners. It can comprise data with the components (molecules, pharmaceutically active components etc.) with their individual concentrations and an overall dosage. This data can be delivered to a composing component or the composing can be integrated into the system.

In line with that, the method according to the present invention can be provided for an automated composing of a customized composition, such as a pharmaceutically active composition. It can comprise steps of inputting genetic data of a plurality of individuals; and inputting time stamped data of the plurality of individuals; combining the genetic data and the time stamped data of each individual to combined data; establishing and/or training a multi-layer neural network on the basis of the combined data; and predicting a disease risks and/or treatment response of the customized composition.

The treatment response can comprise disease data representative to the disease response and/or side effect data representative of positive and/or negative side effects. Also, interactions between the different components can be determined and taken into consideration.

Further, an optimization component can be provided that optimizes positive and/or negative side effects of any treatment and/or combination thereof. This can be integrated into and/or added to the predicting component.

The time stamped data can comprise phenotypic data of individuals. The time stamped data can, as mentioned before and below, comprise health and/or disease relevant data.

A further step of or component for determining the customized composition for an individual by the multi-layer neural network can be present.

Moreover, steps of or components for determining the customized composition for an individual and then predicting a treatment response of the customized composition can be provided.

An arrangement for or step of determining a plurality of potential different customized compositions for an individual and then predicting a treatment response for each customized composition can be provided in line with the present invention.

The treatment response for each customized composition and prioritizing one of the customized compositions considering the disease data and side effect data can be addressed as well. The treatment response can be determined numerically, graphically, in any other kind or combinations thereof.

Also, a determining component for determining the customized composition for an individual by the multi-layer neural network can be present.

The multi-layer network is a deep neural network. The multi-layer network can be a tree shaped deep neural network (TSDNN). The tree-shaped deep neural network (TSDNN) can comprise at least two input layers.

The first input layer can be configured for inputting genetic data and the second input layer for inputting the time stamped data. The inputting can be also pulled automatically from the respective data base(s). The genetic data can comprise at least one of genetic marker data, gender data, age data. There is a number of other data or information that can be made use of, if available. The genetic marker data comprises SNPs representing at least one genetic predisposition.

The time-stamped data can comprise a time-stamp and a medical history data or other data as discussed before and below. The time-stamped data can comprise time-stamps in time bins or aggregated in time bins. The time bins can be bins of 1 to 10 years, preferably bins of 3 to 7 years, and most preferably a bin of 5 years.

The time-stamped data can comprise event code data, the event code data preferably comprising a plurality of status features. The status features can be derived from the event code by a LSTM sequence-to-sequence (seq-to-seq) architecture.

The processing component and/or the neural network can be configured for concatenating the genetic data and the historic data for each individual. The system can be configured for concatenating after the inputting of the genetic data into the first input layer and the inputting of the time stamped data into the second input layer. The system can be configured for generating concatenating features.

The neural network can comprise a first hidden layer that is configured to be or become connected with the concatenating features.

At least a second hidden layer can be configured for activating by rectified linear units (ReLUs).

The neural network can comprise at least a second hidden layer and a third hidden layer.

The system can be configured for connecting the second and third hidden layers for activating by rectified linear units (ReLUs).

The neural network can further comprise a risk score layer that itself can comprise a risk score log layer or a log risk score layer.

The risk score layer can be connected with a hidden layer that can be the outermost or last layer.

The pharmaceutically active composition can comprise a vaccination.

In the system or method thus established and trained a customized pharmaceutically active composition can be composed or predicted for a new individual on the basis of the neural network established according to any one of the preceding system or below embodiments. The present invention also refers to a use of the system according to any of the preceding system or method embodiments for carrying out the method according to any of the preceding method embodiments.

The present invention also covers a computer-related product for carrying out the method according to any of the preceding method embodiments.

The invention bears the preferred advantage to fit a continuously parametric function of time-varying effects that is represented by the deep neural network, instead of a stepwise function of time-varying effects, e.g., as applied in the RNN-SURV model.

The invention is further described with the following numbered embodiments.

Below, system embodiments will be discussed. These embodiments are abbreviated by the letter "S" followed by a number. Whenever reference is herein made to "system embodiments", these embodiments are meant.

S1. System for generating a customized treatment comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with treatment data comprising a plurality of treatments; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a machine learning component configured for training a predicting component for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

S2. System for generating a customized treatment comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with treatment data comprising a plurality of treatments; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a predicting component having been trained by a machine learning component, the predicting component being configured for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

S3. System for simulating a customized treatment comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with treatment data comprising a plurality of treatments; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a machine learning component configured for training a predicting component for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

S4. System for simulating a customized treatment comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with treatment data comprising a plurality of treatments; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a predicting component trained by a machine learning component, the predicting component being configured for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

S5. System for predicting and/or diagnosing a disease comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (SOB) with time-stamped data of the plurality of individuals; c. a third data base (30C) with disease data comprising a plurality of diseases; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a machine learning component configured for training a predicting component for predicting a disease.

S6. System for predicting and/or diagnosing a disease comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with disease data comprising a plurality of diseases; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a predicting component having been trained by a machine learning component, the predicting component being configured for predicting a disease.

S7. System for simulating a customized treatment comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with treatment data comprising a plurality of treatments; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a machine learning component configured for training a predicting component for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

S8. System for simulating a customized treatment comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with treatment data comprising a plurality of treatments; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a predicting component having been trained by a machine learning component, the predicting component being configured for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment. S9. System according to any one of the preceding system embodiments wherein the predicting component is further configured to generate and/or suggest a treatment on the basis of the machine learning.

S10. System according to any one of the preceding system embodiments wherein the treatment data comprises a plurality of ingredients for a plurality of pharmaceutically active compositions.

S11. System according to any one of the preceding system embodiments wherein the treatment data comprises a plurality of pharmaceutically active components and/or compositions.

S12. System according to any one of the preceding system embodiments wherein the treatment data comprises a plurality of dosages and/or combinations of a plurality of customized pharmaceutically active components and/or compositions.

S13. System according to any one of the preceding system embodiments wherein the treatment data comprises one or more electro-magnetic treatment(s).

S14. System according to any one of the preceding system embodiments wherein the treatment data comprises one or more irradiation treatment(s).

S15. System according to any of the preceding system embodiments wherein the machine learning component further comprises a composition component that is configured to compose a customized composed treatment.

S16. System according to any one of the preceding system embodiments wherein the treatment response comprises disease data representative to the disease response and/or side effect data representative of positive and/or negative side effects.

S17. System according to any one of the preceding system embodiments further comprising an optimization component that optimizes positive and/or negative side effects of any treatment and/or combination thereof.

S18. System according to any one of the preceding system embodiments wherein the time stamped data comprises phenotypic data of the individuals. S19. System according to any one of the preceding system embodiments wherein the time stamped data comprises epigenetic data of the individuals.

S20. System according to any one of the preceding system embodiments wherein the time stamped data comprises germline data of the individuals.

S21. System according to any one of the preceding system embodiments wherein the time stamped data comprises somatic data of the individuals, such as aneuploidy, translocation, insertion, deletion, indels and/or SNVs of the individuals.

S22. System according to any one of the preceding system embodiments wherein the time stamped data comprises health and/or disease relevant data of the individuals.

S23. System according to any one of the preceding system embodiments wherein the machine learning network comprises at least a neural network, a deep neural network, a random forest, a random decision tree forest, a deep belief network, a gradient boosting machine or any combination thereof.

S24. System according to any one of the preceding system embodiments wherein the machine learning network is further configured for determining the customized composition for an individual by the multi-layer neural network.

S25. System according to any one of the preceding system embodiments further comprising a determining component for determining the customized composition for an individual and then predicting a treatment response of the customized composition.

S26. System according to any one of the preceding system embodiments wherein the determining component is configured for determining a plurality of potential different customized compositions for an individual and then predicting a treatment response for each customized composition.

S27. System according to the preceding system embodiment further comprising an analyzing component for analyzing the treatment response for each customized composition and prioritizing one of the customized compositions considering the disease data and side effect data. S28. System according to the preceding system embodiment wherein the multi-layer network is a deep neural network (DNN).

S29. System according to any one of the preceding system embodiments wherein the multi-layer network is a tree-shaped deep neural network (TSDNN).

S30. System according to the preceding system embodiment wherein the tree-shaped deep neural network (TSDNN) comprises at least two input layers.

S31. System according to any one of the two preceding system embodiments wherein the first input layer is configured for inputting genetic data and the second input layer for inputting the time stamped data.

S32. System according to any one of the preceding system embodiments wherein the genetic data comprises at least one of genetic marker data, gender data, age data.

S33. System according to the preceding system embodiment wherein the genetic marker data comprises SNPs representing at least one genetic predisposition.

S34. System according to any one of the preceding system embodiments wherein the time stamped data comprises a time stamp and a medical history data.

S35. System according to any one of the preceding system embodiments wherein the time stamped data comprises time stamps in time bins.

S36. System according to the preceding system embodiment wherein the time bins are bins of 1 to 10 years.

S37. System according to any of the two preceding system embodiments wherein the time bins are bins of 3 to 7 years.

S38. System according to any of the two preceding system embodiments wherein the time bin is a bin of 5 years.

S39. System according to any one of the preceding system embodiments wherein the time stamped data comprises event code data.

S40. System according to the preceding system embodiment wherein the event code data comprises a plurality of status features. S41. System according to the preceding system embodiment wherein the status features are derived from the event code by a LSTM sequence-to-sequence (seq- to-seq) architecture.

S42. System according to any one of the preceding system embodiments wherein the processing component and/or the neural network is/are configured for concatenating the genetic data and the historic data for each individual.

S43. System according to the preceding system embodiment wherein the system is configured for concatenating after the inputting of the genetic data into the first input layer and the inputting of the time stamped data into the second input layer.

S44. System according to the preceding system embodiment wherein the system is configured for generating concatenating features.

S45. System according to any of the preceding system embodiments wherein in the neural network comprises a first hidden layer that is configured to be or become connected with the concatenating features.

S46. System according to any one of the preceding system embodiments comprising at least a second hidden layer configured for activating by rectified linear units (ReLUs).

S47. System according to any one of the preceding system embodiments, the neural network comprising at least a second hidden layer and a third hidden layer.

S48. System according to the preceding system embodiment wherein the system is configured for connecting the second and third hidden layers by activating by rectified linear units (ReLUs).

S49. System according to any one of the preceding system embodiments, the neural network further comprising a risk score layer.

S50. System according to the preceding system embodiment wherein the risk score comprises and/or is composed of a plurality of risk factors.

S51. System according to the preceding system embodiment wherein the risk score layer comprises a risk score log layer. S52. System according to any one of the preceding system embodiments further comprising a composing component configured for composing the risk score by a plurality of risk factors.

S53. System according to any of the two preceding system embodiments wherein the risk score layer is connected with a hidden layer.

S54. System according to the preceding system embodiment the risk score layer is connected with the last hidden layer.

S55. System according to any one of the preceding system embodiments wherein the pharmaceutically active composition is a vaccination.

S56. System according to any one of the preceding system embodiments wherein the different kinds of data, such as the genetic data, the time-stamped data and the component data are stored on different storage units (30).

S57. System according to the preceding system embodiment wherein privacy-relevant data, such as the individual data or the link data to an individual, is stored in encrypted form on a storage unit (30).

S58. System according to the two preceding system embodiments wherein the different storage units are separated and at least one of them remotely.

S59. System according to any one of the preceding system embodiments wherein the processing component comprises a plurality of processing devices.

S60. System according to any one of the preceding system embodiments wherein the processing component is located remotely.

S61. System according to any one of the preceding system embodiments wherein the processing component and/or the storage units are cloud-based.

S62. System for an automated generation of a customized treatment for an individual on the basis of the prediction component trained according to any one of the preceding system embodiments.

S63. System for an automated generation of a personalized treatment for an individual on the basis of the prediction component trained according to any one of the preceding system embodiments. Below, method embodiments will be discussed. These embodiments are abbreviated by the letter "M" followed by a number. Whenever reference is herein made to "method embodiments", these embodiments are meant.

M1. Method for an automated generation of a customized treatment, with the following steps: a. inputting genetic data of a plurality of individuals; b. inputting time stamped data of the plurality of individuals; c. combining the genetic data and the time stamped data of each individual to combined data; d. inputting treatment data comprising a plurality of treatments; e. training a predicting component for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

M2. Method for an automated generation of a customized treatment, with the following steps: a. inputting genetic data of a plurality of individuals; b. inputting time stamped data of the plurality of individuals; c. combining the genetic data and the time stamped data of each individual to combined data; d. inputting treatment data comprising a plurality of treatments; e. selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment by a trained predicting component.

M3. Method for the simulation of a customized treatment with the following steps: a. inputting genetic data of a plurality of individuals; b. inputting time stamped data of the plurality of individuals; c. combining the genetic data and the time stamped data of each individual to combined data; d. inputting treatment data comprising a plurality of treatments; e. training a predicting component for selecting at least one treatment and/or simulating a treatment response on the basis of the combined data and at least one treatment.

M4. Method for the simulation of a customized treatment with the following steps: a. inputting genetic data of a plurality of individuals; b. inputting time stamped data of the plurality of individuals; c. combining the genetic data and the time stamped data of each individual to combined data; d. inputting treatment data comprising a plurality of treatments; e. selecting at least one treatment and/or simulating a treatment response on the basis of the combined data and at least one treatment by a trained predicting component.

M5. Method according to any one of the preceding method embodiments with the further step of generating and/or suggesting a treatment on the basis of the machine learning.

M6. Method according to any one of the preceding method embodiments wherein the treatment data comprises a plurality of ingredients for a plurality of customized pharmaceutically active compositions.

M7. Method according to any one of the preceding method embodiments wherein the treatment data comprises a plurality of pharmaceutically active components and/or compositions.

M8. Method according to any one of the preceding method embodiments wherein the treatment data comprises a plurality of dosages and/or combinations of a plurality of customized pharmaceutically active components and/or compositions.

M9. Method according to any one of the preceding method embodiments wherein the treatment data comprises one or more electro-magnetic treatment(s).

M10. Method according to any one of the preceding method embodiments wherein the treatment data comprises one or more irradiation treatment(s).

M11. Method according to any of the preceding method embodiments wherein the machine learning component further comprises a composition component that is configured to compose a customized composed treatment.

M12. Method according to any one of the preceding method embodiments wherein the treatment response comprises disease data representative to the disease response and/or side effect data representative of positive and/or negative side effects. M13. Method according to any one of the preceding method embodiments further comprising an optimization component that optimizes positive and/or negative side effects of any treatment and/or combination thereof.

M14. Method according to any one of the preceding method embodiments wherein the time stamped data comprises phenotypic data of the individuals.

M15. Method according to any one of the preceding method embodiments wherein the time stamped data comprises epigenetic data of the individuals.

M16. Method according to any one of the preceding method embodiments wherein the time stamped data comprises germline data of the individuals.

M17. Method according to any one of the preceding method embodiments wherein the time stamped data comprises somatic data, such as aneuploidy, translocation, insertion, deletion, indels and/or SNVs of the individuals.

M18. Method according to any one of the preceding method embodiments wherein the time stamped data comprises health and/or disease relevant data of the individuals.

M19. Method according to any one of the preceding method embodiments wherein the machine learning network comprises at least a neural network, a deep neural network, a random forest, a random decision tree forest, a deep belief network, a gradient boosting machine or any combination thereof.

M20. Method according to any one of the preceding method embodiments wherein the customized treatment is a customized composition of a pharmaceutically active composition.

M21. Method according to any one of the preceding method embodiments wherein the customized treatment is a customized electro-magnetic treatment.

M22. Method according to any one of the preceding method embodiments wherein the customized treatment is a customized irradiation treatment.

M23. Method according to any one of the preceding method embodiments wherein the treatment response comprises disease data representative to the disease response and/or side effect data representative of positive and/or negative side effects. M24. Method according to any one of the preceding method embodiments with the further step of determining the customized composition for an individual by the multi-layer neural network.

M25. Method according to any one of the preceding method embodiments with the steps of determining the customized composition for an individual and then predicting a treatment response of the customized composition.

M26. Method according to any one of the preceding method embodiments with the steps of determining a plurality of potential different customized compositions for an individual and then predicting a treatment response for each customized composition.

M27. Method according to the preceding method embodiment with the further step of analyzing the treatment response for each customized composition and prioritizing one of the customized compositions considering the disease data and side effect data.

M28. Method according to any one of the preceding method embodiments wherein the multi-layer network is a deep neural network.

M29. Method according to any one of the preceding method embodiments wherein the multi-layer network is a tree-shaped deep neural network (TSDNN).

M30. Method according to the preceding method embodiment wherein the tree-shaped deep neural network (TSDNN) comprises at least two input layers.

M31. Method according to any one of the two preceding method embodiments wherein the tree-shaped deep neural network (TSDNN) comprises two input layers.

M32. Method according to any one of the two preceding method embodiments wherein the genetic data is input by the first input layer and the time stamped data is input by the second input layer.

M33. Method according to any one of the preceding method embodiments wherein the genetic data comprises at least one of genetic marker data, gender data, age data.

M34. Method according to the preceding method embodiment wherein the genetic marker data comprises SNPs representing at least one genetic predisposition. M35. Method according to any one of the preceding method embodiments wherein the time stamped data comprises a time stamp and a medical history data.

M36. Method according to any one of the preceding method embodiments wherein the time stamped data comprises time stamps in time bins.

M37. Method according to the preceding method embodiment wherein the time bins are bins of 1 to 10 years.

M38. Method according to any of the two preceding method embodiment wherein the time bins are bins of 3 to 7 years.

M39. Method according to any of the two preceding method embodiment wherein the time bin is a bin of 5 years.

M40. Method according to any one of the preceding method embodiments wherein the time stamped data comprises event code data.

M41. Method according to the preceding method embodiment wherein the event code data comprises a plurality of status features.

M42. Method according to the preceding method embodiment wherein the status features are derived from the event code by a LSTM sequence-to-sequence (seq- to-seq) architecture.

M43. Method according to any one of the preceding method embodiments wherein the step of combining is a step of concatenating the genetic data and the historic data for each individual.

M44. Method according to the preceding method embodiment wherein the concatenating step is performed after the inputting of the genetic data into the first input layer and the inputting of the time stamped data into the second input layer.

M45. Method according to the preceding method embodiment wherein the concatenating step generates concatenating features.

M46. Method according to any of the preceding method embodiments with the further step of connecting a first hidden layer with the concatenating features. M47. Method according to any one of the preceding method embodiments with the further step of connecting at least a second hidden layer by activating by rectified linear units (ReLUs).

M48. Method according to any one of the preceding method embodiments with the further step of connecting at least a second hidden layer and a third hidden layer by activating by rectified linear units (ReLUs).

M49. Method according to any one of the preceding method embodiments with the further step of providing a risk score layer.

M50. Method according to the preceding method embodiment with the further step of providing the risk score layer as a risk score log layer.

M51. Method according to any one of the preceding method embodiments with the step of composing the risk score by a plurality of risk factors.

M52. Method according to any of the two preceding method embodiments with the further step of connecting the risk score layer with a hidden layer.

M53. Method according to the preceding method embodiment with the further step of connecting the risk score layer with the last hidden layer.

M54. Method according to any one of the preceding method embodiments wherein the pharmaceutically active composition is a vaccination.

M55. Method according to any one of the preceding method embodiments further comprising a component database comprising dosage data for each component.

M56. Method according to the preceding method embodiment wherein the risk factors are associated with the dosage data.

M57. Method according to any one of the preceding 8 method embodiments with the further step of determining the composition on the basis of the risk score.

M58. Method according to any one of the preceding 8 method embodiments with the further step of determining the composition on the basis of the risk score and/or dosage data of each component. M59. Method for an automated generation of a customized treatment for an individual on the basis of prediction component trained according to any one of the preceding method embodiments.

M60. Method for an automated generation of a customized treatment for an individual on the basis of prediction component trained according to any one of the preceding method embodiments.

M61. Method for an automated generation of a personalized treatment for an individual on the basis of prediction component trained according to any one of the preceding method embodiments.

M62. Method for an automated generation of a personalized pharmaceutically active composition for an individual on the basis of prediction component trained according to any one of the preceding method embodiments.

Below, use embodiments will be discussed. These embodiments are abbreviated by the letter "U" followed by a number. Whenever reference is herein made to "use embodiments", these embodiments are meant.

U1. Use of the system according to any of the preceding system embodiments for predicting a disease risks and/or treatment response of the customized treatment by applying the system according to any of the preceding system embodiments.

U2. Use of the method according to any of the preceding method embodiments for predicting a disease risks and/or treatment response of the customized treatment by carrying out the method according to any of the preceding method embodiments.

Below, computer related product embodiments will be discussed. These embodiments are abbreviated by the letter "C" followed by a number. Whenever reference is herein made to "computer related product embodiments", these embodiments are meant.

C1. A computer related product with a program that is configured for carrying out the method according to any of the preceding method embodiments.

Below, diagnostic embodiments will be discussed. These embodiments are abbreviated by the letter "D" followed by a number. Whenever reference is herein made to "diagnostic embodiments", these embodiments are meant. D1. Method of diagnosing a patient for predicting a disease risks and/or treatment response of the customized treatment according to any of the preceding method embodiments.

Below, therapeutic embodiments will be discussed. These embodiments are abbreviated by the letter "T" followed by a number. Whenever reference is herein made to "therapeutic embodiments", these embodiments are meant.

T1. Method of treating a patient with a customized treatment composed according to any of the preceding method embodiments.

The present invention will now be described with reference to the accompanying drawings, which illustrate embodiments of the invention. These embodiments should only exemplify, but not limit, the present invention.

Figure Description

Fig. 1 schematically exemplifies a system hardware architecture in accordance with the present invention;

Fig. 2 schematically exemplifies a risk for the outbreak or suffering from RLS over lifetime of an individual or object and different treatment impacts and the impact of specific treatments;

Fig. 3 shows an example or embodiment of the hardware/software architecture for the training and learning of the neural network;

Fig. 4 shows the performance evaluation according to the present invention with respect to the example of an RLS profiling.

Table 1 lists a summary of RLS relevant SNPS;

Table 2 shows the polygenic risk prediction performance in line with the present invention; and

Table 3 the DNN risk performance.

Description of preferred embodiments as exemplified in the figures

It is noted that not all the drawings carry all the reference signs. Instead, in some of the drawings, some of the reference signs have been omitted for sake of brevity and simplicity of illustration. Embodiments of the present invention will now be described with reference to the accompanying drawings. Fig. 1 provides a schematic view onto a computing device 100. The computing device 100 may comprise a computing unit 35, a first data storage unit 30A, a second data storage unit 30B and a third data storage unit 30C.

The computing device 100 can be a single computing device or an assembly of computing devices. The computing device 100 can be locally arranged or remotely, such as a cloud solution.

On the different data storage units 30 the different data can be stored, such as the genetic data on the first data storage 30A, the time stamped data and/or event code data and/or phenotypic data on the second data storage 30B and privacy sensitive data, such as the connection of the before-mentioned data to an individual, on the thirds data storage 30C.

Additional data storage can be also provided and/or the ones mentioned before can be combined at least in part. Another data storage (not shown) can comprise data specifying the composition or pharmaceutically active composition and/or medication data, such as pharmaceutical activities, side effects, interactions between the different components etc. This data can also be provided on one or more of the before-mentioned data storages.

The computing unit 35 can access the first data storage unit 30A, the second data storage unit 30B and the third data storage unit 30C through the internal communication channel 160, which can comprise a bus connection 160.

The computing unit 30 may be single processor or a plurality of processors, and may be, but not limited to, a CPU (central processing unit), GPU (graphical processing unit), DSP (digital signal processor), APU (accelerator processing unit), ASIC (application-specific integrated circuit), ASIP (application-specific instruction-set processor) or FPGA (field programable gate array). The first data storage unit 30A may be singular or plural, and may be, but not limited to, a volatile or non-volatile memory, such as a random access memory (RAM), Dynamic RAM (DRAM), Synchronous Dynamic RAM (SDRAM), static RAM (SRAM), Flash Memory, Magneto-resistive RAM (MRAM), Ferroelectric RAM (F-RAM), or Parameter RAM (P-RAM).

The second data storage unit 30B may be singular or plural, and may be, but not limited to, a volatile or non-volatile memory, such as a random access memory (RAM), Dynamic RAM (DRAM), Synchronous Dynamic RAM (SDRAM), static RAM (SRAM), Flash Memory, Magneto-resistive RAM (MRAM), Ferroelectric RAM (F-RAM), or Parameter RAM (P-RAM). The third data storage unit 30C may be singular or plural, and may be, but not limited to, a volatile or non-volatile memory, such as a random access memory (RAM), Dynamic RAM (DRAM), Synchronous Dynamic RAM (SDRAM), static RAM (SRAM), Flash Memory, Magneto-resistive RAM (MRAM), Ferroelectric RAM (F-RAM), or Parameter RAM (P-RAM).

It should be understood that generally, the first data storage unit 30A (also referred to as encryption key storage unit 30A), the second data storage unit 30B (also referred to as data share storage unit 30B), and the third data storage unit 30C (also referred to as decryption key storage unit 30C) can also be part of the same memory. That is, only one general data storage unit 30 per device may be provided, which may be configured to store the respective encryption key (such that the section of the data storage unit 30 storing the encryption key may be the encryption key storage unit 30A), the respective data element share (such that the section of the data storage unit 30 storing the data element share may be the data share storage unit 30B), and the respective decryption key (such that the section of the data storage unit 30 storing the decryption key may be the decryption key storage unit 30A).

In some embodiments, the third data storage unit 30C can be a secure memory device 30C, such as, a self-encrypted memory, hardware-based full disk encryption memory and the like which can automatically encrypt all of the stored data. The data can be decrypted from the memory component only upon successful authentication of the party requiring to access the third data storage unit 30C, wherein the party can be a user, computing device, processing unit and the like. In some embodiments, the third data storage unit 30C can only be connected to the computing unit 35 and the computing unit 35 can be configured to never output the data received from the third data storage unit 30C. This can ensure a secure storing and handling of the encryption key (i.e. private key) stored in the third data storage unit 30C.

In some embodiments, the second data storage unit 30B may not be provided but instead the computing device 100 can be configured to receive a corresponding encrypted share from the database 60. In some embodiments, the computing device 100 may comprise the second data storage unit 30B and can be configured to receive a corresponding encrypted share from the database 60.

The computing device 100 may comprise a further memory component 140 which may be singular or plural, and may be, but not limited to, a volatile or non-volatile memory, such as a random access memory (RAM), Dynamic RAM (DRAM), Synchronous Dynamic RAM (SDRAM), static RAM (SRAM), Flash Memory, Magneto-resistive RAM (MRAM), Ferroelectric RAM (F-RAM), or Parameter RAM (P-RAM). The memory component 140 may also be connected with the other components of the computing device 100 (such as the computing component 35) through the internal communication channel 160.

Further the computing device 100 may comprise an external communication component 130. The external communication component 130 can be configured to facilitate sending and/or receiving data to/from an external device (e.g. backup device 10, recovery device 20, database 60). The external communication component 130 may comprise an antenna (e.g. WIFI antenna, NFC antenna, 2G/3G/4G/5G antenna and the like), USB port/plug, LAN port/plug, contact pads offering electrical connectivity and the like. The external communication component 130 can send and/or receive data based on a communication protocol which can comprise instructions for sending and/or receiving data. Said instructions can be stored in the memory component 140 and can be executed by the computing unit 35 and/or external communication component 130. The external communication component 130 can be connected to the internal communication component 160. Thus, data received by the external communication component 130 can be provided to the memory component 140, computing unit 35, first data storage unit 30A and/or second data storage unit 30B and/or third data storage unit 30C. Similarly, data stored on the memory component 140, first data storage unit 30A and/or second data storage unit 30B and/or third data storage unit 30C and/or data generated by the commuting unit 35 can be provided to the external communication component 130 for being transmitted to an external device.

In addition, the computing device 100 may comprise an input user interface 110 which can allow the user of the computing device 100 to provide at least one input (e.g. instruction) to the computing device 100. For example, the input user interface 110 may comprise a button, keyboard, trackpad, mouse, touchscreen, joystick and the like.

Additionally, still, the computing device 100 may comprise an output user interface 120 which can allow the computing device 100 to provide indications to the user. For example, the output user interface 110 may be a LED, a display, a speaker and the like.

The output and the input user interface 100 may also be connected through the internal communication component 160 with the internal component of the device 100. The processor may be singular or plural, and may be, but not limited to, a CPU, GPU, DSP, APU, or FPGA. The memory may be singular or plural, and may be, but not limited to, being volatile or non-volatile, such an SDRAM, DRAM, SRAM, Flash Memory, MRAM, F- RAM, or P-RAM.

The data processing device can comprise means of data processing, such as, processor units, hardware accelerators and/or microcontrollers. The data processing device 20 can comprise memory components, such as, main memory (e.g. RAM), cache memory (e.g. SRAM) and/or secondary memory (e.g. HDD, SDD). The data processing device can comprise busses configured to facilitate data exchange between components of the data processing device, such as, the communication between the memory components and the processing components. The data processing device can comprise network interface cards that can be configured to connect the data processing device to a network, such as, to the Internet. The data processing device can comprise user interfaces, such as:

^■ output user interface, such as: o screens or monitors configured to display visual data (e.g. displaying graphical user interfaces of the questionnaire to the user), o speakers configured to communicate audio data (e.g. playing audio data to the user),

^■ input user interface, such as: o camera configured to capture visual data (e.g. capturing images and/or videos of the user), o microphone configured to capture audio data (e.g. recording audio from the user), o keyboard configured to allow the insertion of text and/or other keyboard commands (e.g. allowing the user to enter text data and/or other keyboard commands by having the user type on the keyboard) and/or trackpad, mouse, touchscreen, joystick - configured to facilitate the navigation through different graphical user interfaces of the questionnaire.

The data processing device can be a processing unit configured to carry out instructions of a program. The data processing device can be a system-on-chip comprising processing units, memory components and busses. The data processing device can be a personal computer, a laptop, a pocket computer, a smartphone, a tablet computer. The data processing device can be a server, either local and/or remote. The data processing device can be a processing unit or a system-on-chip that can be interfaced with a personal computer, a laptop, a pocket computer, a smartphone, a tablet computer and/or user interface (such as the upper-mentioned user interfaces).

As can be seen in Fig. 2, a risk for the outbreak or suffering from RLS over lifetime of an individual or object is depicted. On the basis of the individual genome there appears a certain but rather low risk of the individual when being born. In the example mentioned before there are 218 SNPs in 208 genes that are known to be associated for the RLS disease.

While the prevalence of RLS is rather low in the twenties of the individual it raises considerably with pregnancy of the individual that is shown in the example to happen with the age of approximately mid-twenty. Over the subsequent next 10 years the risk drops slightly until the mid-thirties.

With an increase of the BMI the risk then rises again and increases then with age. However, upon treatment which is preferably an individualized or customized treatment the risk drops very considerably. In the example shown, the risk drops to alevel similar to the level at birth.

In Fig. 2 it is further shown how a treatment a has an impact on the risks score. In the further progress and with increasing age another treatment c would allow the risk to raise again while a treatment c will keep the risk at an almost constant level low.

Upon those findings a proper treatment can be individualized, although the influence factors, such as genetic factors, epigenetic factors and health or life incidents aggregate to a very high complexity.

Fig. 3 provides an example or embodiment of the architecture for the training and learning of the neural network. Fig. 3 generally shows an environmental input layer. There is at least one or a number of Long Short-Term Memories (LSTM) in a preferred recurrent neural network architecture with feedback connections. The LSTMs may comprise cells, input gates and output gates as well as a forget gate. They can be arranged sequentially (seq-to-seq architecture) and can represent health and/or medical history, such as Body Mass Index (BMI) at one or more different time points. Other factors can be directed to pregnancy, diabetes etc. Those factors can be contained in time bins, such as 5 year bins. Those time bins can also vary upon needs and can vary, such as from 1 to 10 years bins, also depending on the kind of factor. An encoder in the seq-to-seq architecture can comprise a number of LSTMs that can result in an encode vector that can have multiple dimensions, depending on the LSTMs and their arrangements to each other. The encoder vector is then fed to a decoder that can also comprise a number of LSTMs and can decode the encoder vector into a number of decoder functions or parameters that are now called features.

The features can after that be used for concatenating them with another layer, such as a genetic input layer.

Fig. 3 also schematically depicts in the lower part the input to be concatenated with the data as mentioned before. The data to be connected or concatenated is primarily genetic data. SNPs is just one option for the genetic data to be fed. Also, genomic data could be fed in. The software would the pick and place the relevant sections or sequences or SNPs that are relevant for a certain disease.

The input can also comprise recurrent data, e.g. transmitted by LSTMs. This genetic input layer will be shown and explained in further detail below.

As is apparent from Fig. 3, the present invention is capable to predict a disease risks and treatment response based on age, health/medical history, and genetic background, etc. This figure exemplifies an overview of the inputs and outputs of the invented neural network which will be used for the prediction. As an example, the neurological disease of the so-called restless leg syndrome (RLS) is used. Inputs include the genotypes of 218 single nucleotide polymorphisms (SNPs) located in 208 genes, health tracks and events, RLS diagnosis and treatments. As outputs the network provides time-dependent RLS risks which refer to the occurrence of RLS if an individual is not affected and to RLS progression if RLS has occurred already.

This can constitute the basis not only for the simulation but also for the generation of individualized treatments.

Fig. 3 shows a more detailed insight into the architecture of one example in accordance with the present invention. The TSDNN can have two input layers, multiple hidden layers, and a score layer and output layer. The genetic input layer (221) can directly input features about genetic markers, sex, and age. The environmental input layer (20,1) inputs event codes, representing medical history such as pregnancy, BMI and diabetes in 5 years bins. 10 status features are derived from the event code by a LSTM seq-to-seq architecture. Genetic and environment feature are then combined by concatenating the genetic input layer (221) with the LSTM decode layer (10). The first hidden layer (1,221+10) is fully connected with the concatenated features. Three more fully connected hidden layers follow (3x "dense"), each layer being activated by a Rectified Linear Units (ReLUs). The risk score log(θ) layer (1,10) is fully connected to the last hidden layer ("dense", 1,3011,10) and subject to linear activation.

In the example shown, a time-varying cost function was constructed to train the 0 as a nonlinear time-varying risk score. The cost function is defined as a L1 regularized partial likelihood function of proportional hazards model,

where β is the vector of all weights in the network that is used to fit a parametric time- varying risk function in the proportional hazards model, θi(ti) is the risk score of a given time (age) with θ(t) = exp(σ(t)) being a function of time, covariates and weights, l is the strength of the L1 regularization | |β| | 1. The target layer provides the event status Ci of an individual i.

Risk score log(θ) can be a pure genome-wide risk score if all event inputs are null.

The weights can be optimized using Adaptive Subgradient Methods for Online Learning and Stochastic Optimization (Adadelta).

The baseline cumulative survival function S_o was estimated with maximum likelihood:

Fig. 4 shows the performance evaluation on real conditions. The improvement of risk prediction achieved by the invention was measured as an increase of the area under the ROC curve, and as a higher odds ratio and higher prediction precision for selected risk quantiles. 5-fold cross validation was used to evaluate the prediction performance. AUC increased to 0.885, as compared to raw PRS (0.672) and logistic regression (0.766).

The TSDNN and CART models for lifetime prediction ("FS-DNN-lifetime" and "FS-CART- lifetime") show a shift to higher sensitivity as compared to other lifetime prediction methods which relates to the age-dependent increase of the RLS risk. The two models can predict most cases (high sensitivity) but erroneously predict some controls as cases (reduced specificity). The latter is explainable by the fact that a considerable number of individuals in the control set will develop the disease in their future lifetime. This phenomenon has little impact on the 5-years prediction of the TSDNN which has an excellent time-varying performance while 5-years prediction of the CART method is insufficient.

Table 1 shows actually known SNPs relevant for the RLS disease.

The TSDNN-predicted top 5% of the population has a much larger risk than the subset of individuals that was predicted to represent the top 5% risk group by the raw PRS. The risks were 20.9-fold vs 6.15-fold, respectively (see also tables 2,3).

Assuming that a 90% prediction precision is clinically useful, 10% of the population can be predicted usefully by the TSDNN, while only 0.5% can be predicted usefully by the raw PRS (see table 2,3).

Reference numbers and letters appearing between parentheses in the claims, identifying features described in the embodiments and illustrated in the accompanying drawings, are provided as an aid to the reader as an exemplification of the matter claimed. The inclusion of such reference numbers and letters is not to be interpreted as placing any limitations on the scope of the claims.

The term "at least one of a first option and a second option" is intended to mean the first option or the second option or the first option and the second option.

Whenever a relative term, such as "about", "substantially" or "approximately" is used in this specification, such a term should also be construed to also include the exact term. That is, e.g., "substantially straight" should be construed to also include "(exactly) straight".

Whenever steps were recited in the above or also in the appended claims, it should be noted that the order in which the steps are recited in this text may be accidental. That is, unless otherwise specified or unless clear to the skilled person, the order in which steps are recited may be accidental. That is, when the present document states, e.g., that a method comprises steps (A) and (B), this does not necessarily mean that step (A) precedes step (B), but it is also possible that step (A) is performed (at least partly) simultaneously with step (B) or that step (B) precedes step (A). Furthermore, when a step (X) is said to precede another step (Z), this does not imply that there is no step between steps (X) and (Z). That is, step (X) preceding step (Z) encompasses the situation that step (X) is performed directly before step (Z), but also the situation that (X) is performed before one or more steps (Y1), followed by step (Z). Corresponding considerations apply when terms like "after" or "before" are used.

Claims

1. System for simulating and/or generating a customized treatment comprising: a. a first data base (30A) with genetic data of a plurality of individuals; b. a second data base (30B) with time-stamped data of the plurality of individuals; c. a third data base (30C) with treatment data comprising a plurality of treatments; d. a processing component (35) configured for combining the genetic data and the time-stamped data of each individual to combined data; e. a machine learning component configured for training a predicting component for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

2. System according to the preceding claim wherein the predicting component is further configured to generate and/or suggest a treatment on the basis of the machine learning.

3. System according to any one of the preceding claims wherein the treatment data comprises a plurality of ingredients for a plurality of customized pharmaceutically active compositions.

4. System according to any one of the preceding system embodiments wherein the treatment data comprises a plurality of pharmaceutically active components and/or compositions.

5. System according to any one of the preceding system embodiments wherein the treatment data comprises a plurality of dosages and/or combinations of a plurality of customized pharmaceutically active components and/or compositions.

6. System according to any one of the preceding claims wherein the treatment data comprises one or more electro-magnetic treatment(s).

7. System according to any one of the preceding claims wherein the treatment data comprises one or more irradiation treatment(s).

8. System according to any of the preceding claims wherein the machine learning component further comprises a composition component that is configured to compose a customized composed treatment.

9. System according to any one of the preceding claims wherein the treatment response comprises disease data representative to the disease response and/or side effect data representative of positive and/or negative side effects.

10. System according to any one of the preceding claims wherein the time stamped data comprises epigenetic data of the individuals.

11. System according to any one of the preceding claims wherein the time stamped data comprises germline data of the individuals.

12. System according to any one of the preceding claims wherein the time stamped data comprises somatic data of the individuals, such as aneuploidy, translocation, insertion, deletion, indels and/or SNVs of the individuals.

13. System according to any one of the preceding claims wherein the time stamped data comprises health and/or disease relevant data of the individuals.

14. System according to any one of the preceding claims wherein the machine learning network comprises at least a neural network, a deep neural network, a random forest, a random decision tree forest, a deep belief network, a gradient boosting machine or any combination thereof.

15. Method for an automated simulation or generation of a customized treatment, with the following steps: a. inputting genetic data of a plurality of individuals; b. inputting time stamped data of the plurality of individuals; c. combining the genetic data and the time stamped data of each individual to combined data; d. inputting treatment data comprising a plurality of treatments; e. training a predicting component for selecting at least one treatment and/or predicting a treatment response on the basis of the combined data and at least one treatment.

16. Method according to any one of the preceding claims with the further step of generating and/or suggesting a treatment on the basis of the machine learning.

17. Method according to any one of the preceding claims wherein the treatment data comprises a plurality of ingredients for a plurality of customized pharmaceutically active compositions.

18. Method according to any one of the preceding claims wherein the treatment data comprises a plurality of applications for a plurality of customized pharmaceutically active compositions.

19. Method according to any one of the preceding claims wherein the treatment data comprises one or more electro-magnetic treatment(s).

20. Method according to any one of the preceding claims wherein the treatment data comprises one or more irradiation treatment(s).

21. Method according to any one of the preceding method embodiments wherein the treatment data comprises a plurality of pharmaceutically active components and/or compositions.

22. Method according to any one of the preceding method embodiments wherein the treatment data comprises a plurality of dosages and/or combinations of a plurality of customized pharmaceutically active components and/or compositions.

23. Method according to any one of the preceding claims wherein the treatment response comprises disease data representative to the disease response and/or side effect data representative of positive and/or negative side effects.

24. Method according to any one of the preceding claims wherein the time stamped data comprises epigenetic data of the individuals.

25. Method according to any one of the preceding claims wherein the time stamped data comprises germline data of the individuals.

26. Method according to any one of the preceding claims wherein the time stamped data comprises somatic data of the individuals, such as aneuploidy, translocation, insertion, deletion, indels and/or SNVs of the individuals.

27. Method according to any one of the preceding claims wherein the time stamped data comprises health and/or disease relevant data of the individuals.

28. Method according to any one of the preceding claims wherein the machine learning network comprises at least a neural network, a deep neural network, a random forest, a random decision tree forest, a deep belief network, a gradient boosting machine or any combination thereof.

29. Method according to any one of the preceding claims wherein the customized treatment is a customized composition of a pharmaceutically active composition.

30. Method according to any one of the preceding claims wherein the customized treatment is a customized electro-magnetic treatment.

31. Method according to any one of the preceding claims wherein the customized treatment is a customized irradiation treatment.

32. Method according to any one of the preceding claims wherein the treatment response comprises disease data representative to the disease response and/or side effect data representative of positive and/or negative side effects.

33. Use of the system according to any of the preceding system claims for simulating and/or generating a disease risks and/or treatment response of the customized composition by applying the system according to any of the preceding system claims.

34. Use of the method according to any of the preceding method claims for simulating and/or generating a disease risk and/or treatment response of the customized composition by carrying out the method according to any of the preceding method claims.

35. A computer related product for carrying out the method according to any of the preceding method claims.