CN107066833A - Immunological network categorizing system and method with cell differentiation is oriented to based on fictitious force - Google Patents
Immunological network categorizing system and method with cell differentiation is oriented to based on fictitious force Download PDFInfo
- Publication number
- CN107066833A CN107066833A CN201611061547.8A CN201611061547A CN107066833A CN 107066833 A CN107066833 A CN 107066833A CN 201611061547 A CN201611061547 A CN 201611061547A CN 107066833 A CN107066833 A CN 107066833A
- Authority
- CN
- China
- Prior art keywords
- msub
- mrow
- antibody
- antigen
- fictitious force
- 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.)
- Pending
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
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)
- Peptides Or Proteins (AREA)
Abstract
The invention discloses a kind of immunological network categorizing system and method being oriented to based on fictitious force with cell differentiation, this method comprises the following steps：Step S1, data and immunological network initialization, step S2, t ＜ T or antigen body number of all categories no longer change, step S3, selection antigen g, step S4, the affinity for calculating g and all antibody, step S5, affinity highest antibody classification it is different from g, step S6, selection affinity highest ξ antibody, step S7, the antibody evolution strategy based on fictitious force, step S8, it is adjusted to rational position, step S9, the antibody evolution strategy based on cell differentiation, step S10, immunological network suppression etc..The present invention can be by training effective antibody cell group to lift the accuracy of classification results, effectively the evolutionary process for antibody is instructed, the possibility of wrong identification foreign peoples's sample is reduced, the classification accuracy of antibody cell is improved, with preferable classification performance.
Description
Technical field
It is more particularly to a kind of to be oriented to based on fictitious force and thin the present invention relates to a kind of immunological network categorizing system and method
The immunological network categorizing system and method for born of the same parents' differentiation.
Background technology
Artificial immune network refers to a kind of method that the memory cell for being capable of covering problem space is trained according to antigen, this
In the problem of space just refer to feature space belonging to sample information, even sample data is bivector type, then problem space
Exactly include the two dimensional sample space of antigen sample set.
Earliest artificial immune network model is aiNet (artificial immune system), is described in detail below step：
Step 1: to each antigen：First, the affinity of antibody randomly generated is calculated, the antibody of n high-affinity is chosen；
2nd, n high-affinity antibody produces clonal antibody collection D according to affinity, and for each antibody, its affinity is higher, clones number
It is more；3rd, progress affinity maturation direct to D is processed as in D*, i.e. D each antibody and become according to formula C=C-a (C-X)
Different, affinity is higher, and aberration rate is smaller；Wherein C is network cell matrix, and X is antigen matrix, and a is learning rate or maturing rate,
Set according to Ag-Ab affinity, affinity is higher, and a is smaller；4th, the affinity of each antibody in antigen and D* is obtained；5th, from
A certain proportion of antibody with high-affinity is selected in D*, clone's memory cell is put into and concentrates；6th, obtain memory and concentrate antibody
Similarity, carry out clone inhibition；7th, will remember concentrates antibody to be stored in total memory antibody collection；
Step 2: seeking the similarity of total memory cell antibody, network suppression is carried out；
Step 3: immunological network antibody tormation；
Step 4: end condition：When antibody cell collection reaches specified book, or reach default maximum iteration.
AiNet immune network models are the clustering for data earliest, afterwards, and part research has been carried out accordingly to it
Ground extends to realize the classification to data sample information.
At present, it is the AINC algorithms of the propositions such as Liu Ruochen with the present invention most close artificial immune network sorting technique,
The key technology of this method mainly includes data prediction, netinit, antibody is affine function, network mechanism of proliferation, hypermutation
Different and network competition mechanism.
Data prediction, the purpose of data prediction is in order that each dimensional feature of data occupies identical in classification processing
Proportion.Its specific data prediction is following (11)：
Wherein x represents the pending data read in, and Min (x), Max (x) refer to the maximum in being arranged belonging to some data respectively
And minimum value.
The initialization of immunological network, the initialization of immunological network includes generation and the weight matrix of initial antibodies cell mass
Produce.The average of training sample of all categories is mainly taken as initial antibodies cell；Assuming that pending data set will be divided into C
Class, then W is exactly a C rank square formation, i.e., whereinB is the value and b >=1 being previously set.
The affinity function of antibody affinity function, antibody and antigen uses following formula (12) such as to carry out：
Wherein KiRepresent the number of the i-th class training sample, improved purpose be in order to improve the priority of this classification antigen,
prijRepresent to BiThe correct identification number of jth class training sample, K represents all training sample numbers.
Network mechanism of proliferation, after somatic cells are resisted first by the propagation scale increment set in advance, resists i-th
Body cell BiProliferating population is designated as Mi, pg is that network breeds scale, then in each iteration all to MiIt is updated, evolution
Excellent individual replaces the minimum antibody cell of original affinity in clone afterwards.
Hyper mutation, mainly computing with words two parts between variation and cell constitute.Computing with words between cell is
Refer to MCiIn all antibody cells each dimensional feature value within occur random exchange.En represents MCi(j) it is corresponding reinforcing antigen or
Person is and MCi(j) generic antigen is (when the random number of generation is less than or equal to pmWhen, for reinforcing antigen；Otherwise, it means that after
Person), n is category label, n=1,2 ..., C, the variation formula such as following formula (13) of j-th of antibody cell, (14)：
MCi(j)=MCi(j)-a(MCi(j)-En(k))......(13)
A=srand | | MCi(j)-En(k)||·pm......(14)
Wherein pmFor mutation probability, s is an auto-adaptive parameter, En (k) represent a selected reinforcing antigen or
It is selected and the generic antigen of antibody cell, k=| randL |, | x | represent to being rounded under x, L is represented in En
Comprising antigen number.
With reference to above-mentioned key technology, the specific implementation and step of AINC methods are as follows：
Step 5: reading sample, sample data is pre-processed, the classification number selection training sample according to needed for classification
Area, if class categories number is C, chooses each training fields of C；
Step 6: initialization antibody cell and weight matrix W；
Step 7: network evolution；Tenth, calculating antibody BiAffinity, choose correspondence BiIntensive training sample (reinforcing is anti-
It is former) Agoi, iteration=0；11, network is bred, to BiReplicated by propagation scale pg, generate BiCell pool Mi,
Then i=i+1；Return to 3.1, End if；12, by clone sizes to MiIn individual by affinity size carry out
Antibody cell after clone, clone is designated as MCi；13, to MCiHyper mutation is carried out, MC is obtainedi'；14, to MCi' in it is each
Antibody cell calculates its affinity and its corresponding reinforcing antigen；15, MC is selectedi' in affinity it is high individual to itself
The antibody of low affinity is updated；
Step 8: output network, is classified.
AINC methods mainly use hyper mutation method when selection builds the immunological network for classifying, while using
Be thought that a classification is only recognized by antibody cell.As above two ways has certain limitation, tool
Body surface is now as follows：17, in the case where problem space is complex, a class label is concluded only with an antibody
Situation when it is inaccurate, higher rate of false alarm will be caused, i.e., misrepresented deliberately the classification of sample for other class labels；18, adopt
The mode evolved with random variation is likely to result in immunological network and builds time-consuming higher defect.
The content of the invention
The technical problems to be solved by the invention are to provide a kind of immunological network being oriented to based on fictitious force with cell differentiation
Categorizing system and method, it can be effectively directed to by training effective antibody cell group to lift the accuracy of classification results
The evolutionary process of antibody is instructed, and reduces the possibility of wrong identification foreign peoples's sample, improves the classification accuracy of antibody cell,
With preferable classification performance.
The present invention is to solve above-mentioned technical problem by following technical proposals：One kind is oriented to and cell based on fictitious force
The immunological network categorizing system of differentiation, it is characterised in that including：
Data and immunological network are initialized by initialization module；
Computing module, calculates the affinity of antigen and all antibody；
Immunological network suppression module, network suppression is carried out to the collection of antibodies that training is obtained, and removes affine in set M spend
High antibody cell；
Selecting module, selects all antibody.
Preferably, the initialization module reads sample data set, normalization sample attribute, loading training sample set.
Preferably, in the computing module set of computations each antibody and antigen affinity.
The present invention also provides a kind of immunological network sorting technique being oriented to based on fictitious force with cell differentiation, and it includes following
Step：
Step S1, data and immunological network initialization, read sample data set S, normalization sample attribute, loading training
Sample set G, then test set is T=S-G.Initialize antibody collectionFor each classification of sample, randomly select in G
One antigen adds set M as initial antibodies, initializes iterations t=1, goes to step S2；
Step S2, t ＜ T or antigen body number of all categories no longer change, and iterations t reaches default maximum iteration
T, or (once training refer in G that all antigens are trained finish) antibody number of all categories is no longer sent out twice in training process
Changing, is to go to step S13, otherwise goes to step S3；
Step S3, selection antigen g, go to step S4；
The affinity of each antibody and g, goes to step S5 in step S4, the affinity for calculating g and all antibody, set of computations M；
Step S5, affinity highest antibody classification it is different from g, if affinity highest antibody generic is different from g,
Set M then is added using g as the antibody newly produced, S4 is gone to step, otherwise goes to step S6；
Step S6, selection affinity highest ξ antibody, ξ are that antibody selects number, go to step S7；
Step S7, the antibody evolution strategy based on fictitious force, go to step S8；
Step S8, rational position is adjusted to, is to go to step S10, otherwise goes to step S9；
Step S9, the antibody evolution strategy based on cell differentiation, go to step S10；
Step S10, immunological network suppress, and network suppression is carried out to the collection of antibodies M that training is obtained, and remove parent in set M
With spend high antibody cell, go to step S11；
Step S11, all antibody are chosen, and are to go to step S12, are otherwise gone to step S3；
After the completion of step S12, t=t+1, a training process, by each antibody b in set MiFictitious force place reset
For sky, that is, setIterations t=t+1, goes to step S2；
Step S13, immunological network training terminate.
Preferably, the step S6, it is considered to each antibody b, Z is added to by gbIn, it comprises the following steps：
If step S14, b meets such as following formula, it need not be adjusted；
If step S15, the unit moving step length m being unsatisfactory for as obtained by following formula are adjusted to b, if the b' that adjustment is obtained expires
The mobile condition of convergence of foot, b' is substituted for by set M antibody b, and antibody cell differentiation is otherwise carried out to b operates.
Preferably, the step S7, the antibody evolution strategy based on fictitious force be mainly according to antibody recognized it is neighbouring
Antigen sample, it is located at position in problem space to guiding antibody repair, its is reached more suitable position, to reduce to different
The discrimination of class sample, and then the purpose of improvement antibody classification performance is reached, such antibody evolution strategy will improve each antibody
The performance of grader, and then improve the accuracy of final classification result on the whole.
Preferably, the step S9, during antibody A b movement adjustment, it is understood that there may be special situation causes its nothing
Method meets the mobile condition of convergence, i.e., can not meet the condition of convergence of above formula, and analysis reason mainly includes at following 2 points：22,
Unit motion-vector m values under being oriented to due to fictitious force are excessive, cause it to miss most rational movement in moving process
Restrain position；23, in antibody A b fictitious force place, the affinity between closest cognate antigen and closest foreign peoples's antigen
It is higher, cause antibody A b to be difficult to converge to suitable position, for above-mentioned two problems, algorithm introduces antibody cell differentiation plan
Slightly, its specific method following steps：
32, when antibody A b can not still meet the mobile condition of convergence after repeatedly mobile adjustment, in its fictitious force
Place ZAbMiddle selection and foreign peoples's antigen A g closest AbiAdded as newborn antibody in antibody cell group；
33, by fictitious force place ZAbIn AgiRemove.
Preferably, the antigen A g=＜ [x1,x2,...,xn], C ＞ represent a sample to be sorted, and wherein n is represented
The attribute dimensions of sample, C refers to the class label corresponding to the antigen, xiFor the i-th dimension attribute spy formed after normalization
Value indicative, specific method for normalizing such as following formula：
Wherein Min (xi)、Max(xi) refer to the maximum and minimum value of i-th dimension attribute in all samples respectively.
Preferably, the antibody A b represents that is used for a grader for sample classification, using such as following formula：
Ab=＜ [x1,x2,...,xn], C ＞
Wherein [x1,x2,...,xn] characterize antibody vector characteristics, C is the class label corresponding to the antibody.
Preferably, the affinity, circular such as following formula：
Wherein | | Ab-Ag | | the Euclid distances between antibody A b and antigen A g are represented, n is vector characteristics dimension.
The positive effect of the present invention is：The present invention can be realized to the effective of antibody cell by antigenic information
Evolve, can effectively instruct the evolutionary process of classification antibody cell, that is, improving antibody to the same of similar specimen discerning probability
When, the possibility of its wrong identification foreign peoples's sample is reduced, that is, improves the classification accuracy of antibody cell；In high dimensional data test set
On classification results show that the present invention more balancedly can classify to sample space, and higher classification standard can be reached
True rate；With faster convergence rate, for same data set, required run time is relatively short；With preferable classification
Performance.
Brief description of the drawings
Fig. 1 is schematic flow sheet of the invention.
Embodiment
Present pre-ferred embodiments are provided below in conjunction with the accompanying drawings, to describe technical scheme in detail.
It is oriented to based on fictitious force and the immunological network categorizing system of cell differentiation includes：
Data and immunological network are initialized by initialization module；
Computing module, calculates the affinity of antigen and all antibody；
Immunological network suppression module, network suppression is carried out to the collection of antibodies that training is obtained, and removes affine in set M spend
High antibody cell；
Selecting module, selects all antibody.
The initialization module reads sample data set, normalization sample attribute, loading training sample set.
The affinity of each antibody and antigen in the computing module set of computations.
As shown in figure 1, the present invention is oriented to based on fictitious force and the immunological network sorting technique of cell differentiation includes following step
Suddenly：
Step S1, data and immunological network initialization, read sample data set S, normalization sample attribute, loading training
Sample set G, then test set is T=S-G.Initialize antibody collectionFor each classification of sample, one in G is randomly selected
Individual antigen adds set M as initial antibodies, initializes iterations t=1, goes to step S2；
Step S2, t ＜ T or antigen body number of all categories no longer change, and iterations t reaches default maximum iteration
T, or (once training refer in G that all antigens are trained finish) antibody number of all categories no longer occurs twice in training process
Change, is to go to step S13, otherwise goes to step S3；
Step S3, selection antigen g, go to step S4；
The affinity of each antibody and g, goes to step S5 in step S4, the affinity for calculating g and all antibody, set of computations M；
Step S5, affinity highest antibody classification it is different from g, if affinity highest antibody generic is different from g,
Set M then is added using g as the antibody newly produced, S4 is gone to step, otherwise goes to step S6；
Step S6, selection affinity highest ξ antibody, ξ are that antibody selects number, go to step S7；
Step S7, the antibody evolution strategy based on fictitious force, go to step S8；
Step S8, rational position is adjusted to, is to go to step S10, otherwise goes to step S9；
Step S9, the antibody evolution strategy based on cell differentiation, go to step S10；
Step S10, immunological network suppress, and network suppression is carried out to the collection of antibodies M that training is obtained, and remove parent in set M
With spend high antibody cell, go to step S11；
Step S11, all antibody are chosen, and are to go to step S12, are otherwise gone to step S3；
After the completion of step S12, t=t+1, a training process, by each antibody b in set MiFictitious force place reset
For sky, that is, setIterations t=t+1, goes to step S2；
Step S13, immunological network training terminate.
The step S6, it is considered to each antibody b, Z is added to by gbIn, it comprises the following steps：
If step S14, b meets such as following formula (1), it need not be adjusted；
If step S15, the unit moving step length m being unsatisfactory for as obtained by following formula (2) are adjusted to b, if what adjustment was obtained
B' meets the mobile condition of convergence, and set M antibody b is substituted for into b', and antibody cell differentiation is otherwise carried out to b operates.
The step S7, the antibody evolution strategy based on fictitious force is mainly the neighbouring antigen sample recognized according to antibody
This, it is located at position in problem space to guiding antibody repair, its is reached more suitable position, to reduce to foreign peoples's sample
Discrimination, and then reach and improve the purpose of antibody classification performance, such antibody evolution strategy will improve each antibody classification device
Performance, and then on the whole improve final classification result accuracy, wherein：
The fictitious force place of antibody, for antibody A b, its fictitious force place ZAbIt is made up of its antigen recognized, wherein wrapping
Containing the antigen consistent with Ab classifications, also comprising the inconsistent antigen of classification therewith, therefore, antibody A b such as following formulas (3) are redefined；
Ab=＜ [x1,x2,...,xn],C,ZAb＞ ... (3)
The fictitious force active force of antibody, for antibody A b, definition belongs to ZAbAntigen A giTo the virtual effect produced by Ab
PowerSuch as following formula (4), P is used for the direction for characterizing virtual active force, and when antigen concrete class is consistent with antibody isotype, it takes
It is worth for 1, virtual active force shows as attraction force, otherwise P is -1, and virtual active force shows as repulsion force；
Antibody A b fictitious force place, is broadly divided into the following two kinds situation：20：Work as ZAbIn all antigens and antibody institute
When the classification of category is identical, show that rational position or the antibody of the antibody in immunological network are center antibody, it is relatively defined
The classification to training antigen really is realized, in the absence of the situation of erroneous judgement；
If 21, ZAbIt is middle exist the antigen different from antibody generic, illustrate the antibody exist erroneous judgement antigen can
Can, it is therefore desirable to antibody Ab vector characteristics are adjusted, according to ZAbVirtual effect to antibody Ab makes a concerted effort to draw Ab arrival
Correct position in immunological network, makes it effectively recognize cognate antigen, the possibility of reduction identification foreign peoples's antigen, according to antibody A b
Fictitious force place, define its unit motion-vector m such as following formula (5) in the case where virtual effect makes a concerted effort to guide：
Ab in formulak、Agi,kAntibody A b and antigen A g is represented respectivelyiKth dimensional vector feature, parameter lambda be used for adjust unit shifting
Moving vector m size, to control to be set to 1000 in the one step size that antibody is moved, text.
The direction of fictitious force, due to antigen F and antibody A distance relatively far away from, its by by closer to comparator antibody I
Recognized, and be introduced into A fictitious force place, thus antibody cell A is by by cognate antigen E attraction fEAWith foreign peoples's antigen
C repulsive force fCA, both form the F that makes a concerted effortA, FAA will be guided while close to cognate antigen E, more away from foreign peoples's antigens c,
Memory cell B fictitious force place is mainly made up of antigens c, antigen D and antigen E, and the fictitious force suffered by it makes a concerted effort to be FB, to sentence
Whether the vector characteristics for determining antibody A b are adjusted to rational position, it is necessary to move state convergence to the Ab in mobile status
Judgement, if reasonable antibody after being adjusted is Ab', then it should be met such as the mobile condition of convergence of above formula (1), parameter in formula
θ is mobile convergence factor, is mainly used in judging whether antibody cell is adjusted to suitable position under the traction of fictitious force, for
Different types of sample data set, θ optimal value is also not quite similar, and it depends primarily on foreign peoples's sample information in data set
Difference degree.
Direction after the adjustment of fictitious force, under the guiding of fictitious force, for original antibody A, B, after fictitious force is guided
Antibody A ', the affinity between B' and cognate antigen the trend of lifting is presented, the distance between increase with foreign peoples's antigen, base
The antibody movement being oriented in fictitious force can preferably improve the classification performance of antibody, and the classification for being favorably improved immunological network is accurate
True rate.
The step S9, during antibody A b movement adjustment, it is understood that there may be special situation causes it not meet
The mobile condition of convergence, i.e., can not meet the condition of convergence of above formula (1), and analysis reason mainly includes at following 2 points：22, due to
Unit motion-vector m values under fictitious force is oriented to are excessive, cause it to miss most rational mobile convergence in moving process
Position；23, in antibody A b fictitious force place, affinity between closest cognate antigen and closest foreign peoples's antigen compared with
Height, causes antibody A b to be difficult to converge to suitable position, for above-mentioned two problems, and algorithm introduces antibody cell differentiation plan
Slightly, its specific method following steps：
32, when antibody A b can not still meet the mobile condition of convergence after repeatedly mobile adjustment, in its fictitious force
Place ZAbMiddle selection and foreign peoples's antigen A g closest AbiAdded as newborn antibody in antibody cell group；
33, by fictitious force place ZAbIn AgiRemove.
The antigen A g=＜ [x1,x2,...,xn], C ＞ represent a sample to be sorted, and wherein n represents the category of sample
Property dimension, C refers to the class label corresponding to the antigen, xiFor the i-th dimension attributive character value formed after normalization, specifically
Method for normalizing such as following formula (6)：
Wherein Min (xi)、Max(xi) refer to the maximum and minimum value of i-th dimension attribute in all samples respectively.
The antibody A b represents that is used for a grader for sample classification, using such as following formula (7)：
Ab=＜ [x1,x2,...,xn], C ＞ ... (7)
Wherein [x1,x2,...,xn] characterize antibody vector characteristics, C is the class label corresponding to the antibody.
The affinity, circular such as following formula (8)：
Wherein | | Ab-Ag | | the Euclid distances between antibody A b and antigen A g are represented, n is vector characteristics dimension.
The present invention can be instructed effectively for the evolutionary process of antibody, by training effective antibody cell group to carry
Rise the accuracy of classification results.
Particular embodiments described above, technical problem, technical scheme and beneficial effect to the solution of the present invention are carried out
It is further described, should be understood that the specific embodiment that the foregoing is only of the invention, is not used to limit
The system present invention, within the spirit and principles of the invention, any modification, equivalent substitution and improvements done etc. should be included in
Within protection scope of the present invention.
Claims (10)
1. a kind of immunological network categorizing system being oriented to based on fictitious force with cell differentiation, it is characterised in that including：
Data and immunological network are initialized by initialization module；
Computing module, calculates the affinity of antigen and all antibody；
Immunological network suppression module, network suppression is carried out to the collection of antibodies that training is obtained, and removes affinity in set M too high
Antibody cell；
Selecting module, selects all antibody.
2. the immunological network categorizing system as claimed in claim 1 being oriented to based on fictitious force with cell differentiation, it is characterised in that
The initialization module reads sample data set, normalization sample attribute, loading training sample set.
3. the immunological network categorizing system as claimed in claim 1 being oriented to based on fictitious force with cell differentiation, it is characterised in that
The affinity of each antibody and antigen in the computing module set of computations.
4. a kind of immunological network sorting technique being oriented to based on fictitious force with cell differentiation, it is characterised in that it includes following step
Suddenly：
Step S1, data and immunological network initialization, read sample data set S, normalization sample attribute, loading training sample
Collect G, then test set is T=S-G；Initialize antibody collectionFor each classification of sample, one randomly selected in G resists
Original work are that initial antibodies add set M, initialize iterations t=1, go to step S2；
Step S2, t ＜ T or antigen body number of all categories no longer change, and iterations t reaches default maximum iteration T, or
It is that antibody number of all categories no longer changes in training process twice, is to go to step S13, otherwise goes to step S3；
Step S3, selection antigen g, go to step S4；
The affinity of each antibody and antigen g, turns step in step S4, the affinity for calculating antigen g and all antibody, set of computations M
Rapid S5；
Step S5, affinity highest antibody classification it is different from antigen g, if affinity highest antibody generic and antigen g
Difference, then add set M using antigen g as the antibody newly produced, go to step S4, otherwise go to step S6；
Step S6, selection affinity highest ξ antibody, ξ are that antibody selects number, go to step S7；
Step S7, the antibody evolution strategy based on fictitious force, go to step S8；
Step S8, rational position is adjusted to, is to go to step S10, otherwise goes to step S9；
Step S9, the antibody evolution strategy based on cell differentiation, go to step S10；
Step S10, immunological network suppress, and carry out network suppression to the collection of antibodies M that training is obtained, remove affinity in set M
Too high antibody cell, goes to step S11；
Step S11, all antibody are chosen, and are to go to step S12, are otherwise gone to step S3；
After the completion of step S12, t=t+1, a training process, by each antibody b in set MiFictitious force place reset to sky,
SetIterations t=t+1, goes to step S2；
Step S13, immunological network training terminate.
5. the immunological network sorting technique as claimed in claim 4 being oriented to based on fictitious force with cell differentiation, it is characterised in that
The step S6, it is considered to each antibody b, Z is added to by gbIn, it comprises the following steps：
If step S14, b meets such as following formula, it need not be adjusted；
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>R</mi>
<mi>s</mi>
</msub>
<mo>=</mo>
<mi>M</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>f</mi>
<mrow>
<mi>a</mi>
<mi>f</mi>
<mi>f</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mi>t</mi>
<mi>y</mi>
</mrow>
</msub>
<mo>(</mo>
<mrow>
<msub>
<mi>Ag</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msup>
<mi>Ab</mi>
<mo>&prime;</mo>
</msup>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>,</mo>
<msub>
<mi>Ag</mi>
<mi>i</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>Z</mi>
<mrow>
<mi>A</mi>
<mi>b</mi>
</mrow>
</msub>
<mo>,</mo>
<msub>
<mi>C</mi>
<mrow>
<msub>
<mi>Ag</mi>
<mi>i</mi>
</msub>
</mrow>
</msub>
<mo>=</mo>
<msub>
<mi>C</mi>
<mrow>
<mi>A</mi>
<mi>b</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>R</mi>
<mrow>
<mi>n</mi>
<mi>s</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>M</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>f</mi>
<mrow>
<mi>a</mi>
<mi>f</mi>
<mi>f</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mi>t</mi>
<mi>y</mi>
</mrow>
</msub>
<mo>(</mo>
<mrow>
<msub>
<mi>Ag</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<msup>
<mi>Ab</mi>
<mo>&prime;</mo>
</msup>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>,</mo>
<msub>
<mi>Ag</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>Z</mi>
<mrow>
<mi>A</mi>
<mi>b</mi>
</mrow>
</msub>
<mo>,</mo>
<msub>
<mi>C</mi>
<mrow>
<msub>
<mi>Ag</mi>
<mi>i</mi>
</msub>
</mrow>
</msub>
<mo>&NotEqual;</mo>
<msub>
<mi>C</mi>
<mrow>
<mi>A</mi>
<mi>b</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>R</mi>
<mrow>
<mi>n</mi>
<mi>s</mi>
</mrow>
</msub>
<mo>=</mo>
<msub>
<mi>&theta;R</mi>
<mi>s</mi>
</msub>
<mo>,</mo>
<mn>0</mn>
<mo><</mo>
<mi>&theta;</mi>
<mo><</mo>
<mn>1</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
1
If step S15, the unit moving step length m being unsatisfactory for as obtained by following formula are adjusted to b, moved if the b' that adjustment is obtained is met
The dynamic condition of convergence, b' is substituted for by set M antibody b, and antibody cell differentiation is otherwise carried out to b operates.
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mi>m</mi>
<mo>=</mo>
<mo>&lsqb;</mo>
<msub>
<mi>m</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>m</mi>
<mn>2</mn>
</msub>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msub>
<mi>m</mi>
<mi>k</mi>
</msub>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msub>
<mi>m</mi>
<mi>n</mi>
</msub>
<mo>&rsqb;</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>m</mi>
<mi>k</mi>
</msub>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>Ag</mi>
<mi>i</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>Z</mi>
<mrow>
<mi>A</mi>
<mi>b</mi>
</mrow>
</msub>
</mrow>
</munder>
<mi>P</mi>
<mo>&times;</mo>
<mfrac>
<mrow>
<msub>
<mi>f</mi>
<mrow>
<mi>a</mi>
<mi>f</mi>
<mi>f</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mi>t</mi>
<mi>y</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>Ag</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>A</mi>
<mi>b</mi>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<msub>
<mi>Ab</mi>
<mi>k</mi>
</msub>
<mo>-</mo>
<msub>
<mi>Ag</mi>
<mrow>
<mi>i</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>&lambda;</mi>
<mo>&times;</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>Ag</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>Z</mi>
<mrow>
<mi>A</mi>
<mi>b</mi>
</mrow>
</msub>
</mrow>
</munder>
<msub>
<mi>f</mi>
<mrow>
<mi>a</mi>
<mi>f</mi>
<mi>f</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mi>t</mi>
<mi>y</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>Ag</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<mi>A</mi>
<mi>b</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mtd>
</mtr>
</mtable>
</mfenced>
6. the immunological network sorting technique as claimed in claim 4 being oriented to based on fictitious force with cell differentiation, it is characterised in that
The step S7, the antibody evolution strategy based on fictitious force is mainly the neighbouring antigen sample recognized according to antibody, and guiding is anti-
It is located at position in problem space for body amendment, its is reached more suitable position, to reduce the discrimination to foreign peoples's sample,
And then the purpose of improvement antibody classification performance is reached, such antibody evolution strategy will improve the performance of each antibody classification device, enter
And improve the accuracy of final classification result on the whole.
7. the immunological network sorting technique as claimed in claim 4 being oriented to based on fictitious force with cell differentiation, it is characterised in that
The step S9, during antibody A b movement adjustment, it is understood that there may be special situation causes it can not meet mobile convergence
Condition, i.e., can not meet the condition of convergence of above formula, and analysis reason mainly includes at following 2 points：22, because fictitious force is oriented to
Under unit motion-vector m values it is excessive, cause it to miss most rational mobile convergence position in moving process；20
3rd, in antibody A b fictitious force place, the affinity between closest cognate antigen and closest foreign peoples's antigen is higher, causes antibody
Ab is difficult to converge to suitable position, for above-mentioned two problems, and algorithm introduces antibody cell differentiation strategy, its specific method
Following steps：
32, when antibody A b can not still meet the mobile condition of convergence after repeatedly mobile adjustment, in its fictitious force place
ZAbMiddle selection and foreign peoples's antigen A g closest AbiAdded as newborn antibody in antibody cell group；
33, by fictitious force place ZAbIn AgiRemove.
8. the immunological network sorting technique as claimed in claim 4 being oriented to based on fictitious force with cell differentiation, it is characterised in that
The antigen A g=＜ [x1,x2,...,xn], C ＞ represent a sample to be sorted, and wherein n represents the attribute dimensions of sample, C
Refer to the class label corresponding to the antigen, xiFor the i-th dimension attributive character value formed after normalization, specific normalization side
Method such as following formula：
<mrow>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mi>M</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>M</mi>
<mi>a</mi>
<mi>x</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>M</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
Wherein Min (xi)、Max(xi) refer to the maximum and minimum value of i-th dimension attribute in all samples respectively.
9. the immunological network sorting technique as claimed in claim 4 being oriented to based on fictitious force with cell differentiation, it is characterised in that
The antibody A b represents that is used for a grader for sample classification, using such as following formula：
Ab=＜ [x1,x2,...,xn], C ＞
Wherein [x1,x2,...,xn] characterize antibody vector characteristics, C is the class label corresponding to the antibody.
10. the immunological network sorting technique as claimed in claim 4 being oriented to based on fictitious force with cell differentiation, its feature is existed
In the affinity, circular such as following formula：
<mrow>
<msub>
<mi>f</mi>
<mrow>
<mi>a</mi>
<mi>f</mi>
<mi>f</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mi>t</mi>
<mi>y</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mi>g</mi>
<mo>,</mo>
<mi>A</mi>
<mi>b</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mn>1</mn>
<mo>-</mo>
<mfrac>
<mrow>
<mo>|</mo>
<mo>|</mo>
<mi>A</mi>
<mi>b</mi>
<mo>-</mo>
<mi>A</mi>
<mi>g</mi>
<mo>|</mo>
<mo>|</mo>
</mrow>
<mi>n</mi>
</mfrac>
</mrow>
Wherein | | Ab-Ag | | the Euclid distances between antibody A b and antigen A g are represented, n is vector characteristics dimension.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611061547.8A CN107066833A (en) | 2016-11-25 | 2016-11-25 | Immunological network categorizing system and method with cell differentiation is oriented to based on fictitious force |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611061547.8A CN107066833A (en) | 2016-11-25 | 2016-11-25 | Immunological network categorizing system and method with cell differentiation is oriented to based on fictitious force |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107066833A true CN107066833A (en) | 2017-08-18 |
Family
ID=59618871
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611061547.8A Pending CN107066833A (en) | 2016-11-25 | 2016-11-25 | Immunological network categorizing system and method with cell differentiation is oriented to based on fictitious force |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107066833A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090285148A1 (en) * | 2008-05-19 | 2009-11-19 | Microsoft Corporation | Natural network coding for multi-hop wireless network |
| CN101794393A (en) * | 2010-01-04 | 2010-08-04 | 西安电子科技大学 | Target identification method of remote sensing image of artificial immune network based on self-adaptive PSO (Particle Swarm Optimization) |
| CN104615679A (en) * | 2015-01-21 | 2015-05-13 | 华侨大学 | Multi-agent data mining method based on artificial immunity network |
-
2016
- 2016-11-25 CN CN201611061547.8A patent/CN107066833A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090285148A1 (en) * | 2008-05-19 | 2009-11-19 | Microsoft Corporation | Natural network coding for multi-hop wireless network |
| CN101794393A (en) * | 2010-01-04 | 2010-08-04 | 西安电子科技大学 | Target identification method of remote sensing image of artificial immune network based on self-adaptive PSO (Particle Swarm Optimization) |
| CN104615679A (en) * | 2015-01-21 | 2015-05-13 | 华侨大学 | Multi-agent data mining method based on artificial immunity network |
Non-Patent Citations (2)
| Title |
|---|
| 刘若辰等: ""一种新的人工免疫网络算法及其在复杂数据分类中的应用"", 《电子与信息学报》 * |
| 苟世宁等: ""一种改进的模糊人工免疫网络数据分类方法"", 《西安交通大学学报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109657584B (en) | Improved LeNet-5 fusion network traffic sign identification method for assisting driving | |
| CN111553193B (en) | Visual SLAM closed-loop detection method based on lightweight deep neural network | |
| CN105975573B (en) | A kind of file classification method based on KNN | |
| CN107169035B (en) | A kind of file classification method mixing shot and long term memory network and convolutional neural networks | |
| CN111931902B (en) | Generating countermeasure network model and vehicle track prediction method using generating countermeasure network model | |
| CN104035996B (en) | Field concept abstracting method based on Deep Learning | |
| CN108090510A (en) | A kind of integrated learning approach and device based on interval optimization | |
| CN112465120A (en) | Fast attention neural network architecture searching method based on evolution method | |
| CN111860171A (en) | Method and system for detecting irregular-shaped target in large-scale remote sensing image | |
| CN107506786A (en) | A kind of attributive classification recognition methods based on deep learning | |
| CN108846047A (en) | A kind of picture retrieval method and system based on convolution feature | |
| CN109902740A (en) | It is a kind of based on more algorithm fusions it is parallel learn Industry Control intrusion detection method again | |
| CN110363230A (en) | Stacking integrated sewage handling failure diagnostic method based on weighting base classifier | |
| CN110532946A (en) | A method of the green vehicle spindle-type that is open to traffic is identified based on convolutional neural networks | |
| CN113032613B (en) | Three-dimensional model retrieval method based on interactive attention convolution neural network | |
| CN109214444B (en) | Game anti-addiction determination system and method based on twin neural network and GMM | |
| CN114841244A (en) | Target detection method based on robust sampling and mixed attention pyramid | |
| CN104751463A (en) | Three-dimensional model optimal visual angle selection method based on sketch outline features | |
| CN115376101A (en) | Incremental learning method and system for automatic driving environment perception | |
| CN114821106A (en) | Cherry tomato detection and identification method based on feature pyramid | |
| CN115984213A (en) | Industrial product appearance defect detection method based on deep clustering | |
| CN115690549A (en) | Target detection method for realizing multi-dimensional feature fusion based on parallel interaction architecture model | |
| CN103149878A (en) | Self-adaptive learning system of numerical control machine fault diagnosis system in multi-agent structure | |
| CN117975090A (en) | Character interaction detection method based on intelligent perception | |
| CN111079840B (en) | Complete image semantic annotation method based on convolutional neural network and concept lattice |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |