# US 2022/0284987 A1 · Prediction device, trained model generation device, prediction method, and trained model generation method

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0284987 A1
- Document ID: US20220284987A1
- Status: Published application
- Kind: A1
- Application number: 17/577,527
- Filed: 2022-01-18
- Published: 2022-09-08
- Earliest claimed: 2021-03-05
- Assignees: Tokyo Institute of Technology
- Applicants: Tokyo Institute of Technology
- Inventors: Yutaka AKIYAMA; Masahito OHUE; Keisuke YANAGISAWA; Yasushi YOSHIKAWA; Jianan LI
- Web page: https://patentgrep.com/patent/US20220284987A1

## CPC

- G16B 15/30: Drug targeting using structural data; Docking or binding prediction
- G16B 35/00: ICT specially adapted for in silico combinatorial libraries of nucleic acids, proteins or peptides
- C07K 7/08: having 12 to 20 amino acids (gastrins C07K14/595; somatostatins C07K14/655; melanotropins C07K14/68)
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G16B 40/20: Supervised data analysis
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- C07K: PEPTIDES (peptides in foodstuffs A23; obtaining protein compositions for foodstuffs, working-up proteins for foodstuffs A23J; preparations for medicinal purposes A61K; peptides containing beta-lactam rings C07D; cyclic dipeptides not having in their molecule any other peptide link than those which form their ring, e.g. piperazine-2,5-diones, C07D; ergot alkaloids of the cyclic peptide type C07D519/02; macromolecular compounds having statistically distributed amino acid units in their molecules, i.e. when the preparation does not provide for a specific; but for a random sequence of the amino acid units, homopolyamides and block copolyamides derived from amino acids C08G69/00; macromolecular products derived from proteins C08H1/00; preparation of glue or gelatine C09H; single cell proteins, enzymes C12N; genetic engineering processes for obtaining peptides C12N15/00; compositions for measuring or testing processes involving enzymes C12Q; investigation or analysis of biological material G01N33/00)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A prediction device extracts each predictive feature vector expressing a feature from a peptide that is a target for biostability prediction. The prediction device generates a predicted value of biostability of the prediction target cyclic peptide by inputting plural predictive feature vectors into a trained model pre-trained to output a predicted value of peptide biostability.

## Claims

### Claim 1

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract a predictive feature vector expressing a feature from a peptide that is a target for biostability prediction;
adjust a length of the predictive feature vector to a prescribed length; and
generate a predicted value of biostability for the prediction target peptide by inputting the predictive feature vector adjusted in length into a trained model pre-trained to output a predicted value of peptide biostability from a feature vector expressing a feature of a peptide.

### Claim 2 (depends on 1)

The prediction device of claim 1, wherein the processor is configured to:
adjust the length of the predictive feature vector by a padding method or by conversion using a linear interpolation method.

### Claim 3

A trained model generation device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract a training feature vector expressing a feature from each of a plurality of training peptides;
adjust a length of each training feature vector for each of the plurality of training peptides to a prescribed length; and
generate a trained model for outputting a predicted value for peptide biostability from a feature vector expressing a feature of a peptide by executing a machine learning algorithm based on training data that is the training feature vectors adjusted in length paired with correct values of biostability for the training peptides.

### Claim 4

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract each predictive feature vector expressing a feature from a cyclic peptide that is a target for biostability prediction for instances in which each of a plurality of residues contained in the cyclic peptide is at a start point of a cyclic sequence; and
generate a predicted value for biostability of the prediction target cyclic peptide by inputting a plurality of predictive feature vectors into a trained model pre-trained to output a predicted value of peptide biostability from a feature vector expressing a feature of a cyclic peptide.

### Claim 5 (depends on 4)

The prediction device of claim 4, wherein the processor is configured to input each of the plurality of predictive feature vectors into the trained model and to generate a representative value of a predicted value for biostability of the prediction target cyclic peptide for each of a plurality of feature vectors output from the trained model.

### Claim 6

A trained model generation device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract a training feature vector expressing a feature from out of a plurality of training cyclic peptides for instances in which each of a plurality of residues contained in the respective training cyclic peptide is at a start point of a cyclic sequence; and
generate a trained model for outputting a predicted value of biostability of a cyclic peptide from a feature vector expressing a feature of a cyclic peptide by executing a machine learning algorithm based on training data that is a plurality of training feature vectors for each of a plurality of training cyclic peptides paired with a correct value of biostability for the respective training cyclic peptide.

### Claim 7

A trained model generation device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract a first training feature vector expressing a feature from each of a plurality of training cyclic peptides;
generate a plurality of second training feature vectors for each of the first training feature vectors by cyclically shifting elements of the first training feature vector, and generate training data expressed by the first training feature vector and the plurality of second training feature vectors paired with a correct value for biostability of the respective training cyclic peptide; and
generate a trained model for outputting a predicted value of biostability of a cyclic peptide from a feature vector expressing a feature of a cyclic peptide by executing a machine learning algorithm based on a plurality of training data.

### Claim 8 (depends on 7)

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract a predictive feature vector expressing a feature from a cyclic peptide that is a target for biostability prediction; and
generate a predicted value of biostability for the prediction target peptide by inputting the predictive feature vector into the trained model generated by the trained model generation device of claim 7.

### Claim 9

A trained model generation device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
generate a trained convolutional neural network model for outputting a predicted value of biostability of a cyclic peptide from a feature vector expressing a feature of a cyclic peptide by, based on training data expressed by a training feature vector expressing a feature extracted from each of a plurality of training cyclic peptides paired with a correct value of biostability for the plurality of respective training cyclic peptides, executing a machine learning algorithm employing a convolutional neural network model including a both-end-adjacency layer in which elements at both ends of the training feature vector are placed adjacent to one another.

### Claim 10

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract a predictive feature vector expressing a feature from a cyclic peptide that is a target for biostability prediction; and
generate a predicted value of biostability of the prediction target peptide by inputting the predictive feature vector into a trained convolutional neural network model including a both-end-adjacency layer in which elements at both ends of a feature vector expressing a feature of a cyclic peptide are placed adjacent to one another, the trained convolutional neural network model being configured to output a predicted value of biostability of a peptide from the feature vector.

### Claim 11

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
generate a plurality of conformations adoptable by a peptide that is a target for biostability prediction;
select a conformation to be subjected to docking calculation from out of the plurality of conformations based on a prescribed selection criteria; and
predict an biostability of the prediction target peptide by performing docking calculation between the prediction target peptide corresponding to the conformation and a blood plasma protein.

### Claim 12 (depends on 11)

The prediction device of claim 11, wherein the processor is configured to select the conformation to be subjected to docking calculation from the plurality of conformations based on at least one factor from out of:
a length of a side chain of the prediction target peptide when adopting the conformation;
a straightness of a side chain of the prediction target peptide when adopting the conformation;
a structure of a root portion of a side chain of the prediction target peptide when adopting the conformation;
a three dimensional shape of a vicinity of a leading end portion of a side chain of the prediction target peptide when adopting the conformation; or
a physical condition representing presence or absence of a charged atom contained in a side chain of the prediction target peptide when adopting the conformation.

### Claim 13

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
compute a predicted value of a first biostability expressing an biostability of a peptide that is a target for biostability prediction by performing docking calculation between the peptide and a blood plasma protein;
generate a predicted value of a second biostability expressing an biostability of the peptide by inputting a feature vector extracted from the prediction target peptide into a trained model generated in advance by a machine learning algorithm; and
compute an biostability of the peptide by consolidating the first biostability predicted value with the second biostability predicted value.

### Claim 14

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
compute a docking profile including a docking score between a peptide that is a target for biostability prediction and a blood plasma protein by performing docking calculation between the peptide and the blood plasma protein; and
generate a predicted value of biostability of the prediction target peptide by inputting a predictive feature vector including the docking profile into a trained model generated in advance by a machine learning algorithm.

### Claim 15 (depends on 14)

The prediction device of claim 14, wherein the docking profile includes at least one of:
a docking score between the peptide and each residue in a pocket of the blood plasma protein; or
an overall docking score between the peptide and the blood plasma protein.

### Claim 16 (depends on 14)

The prediction device of claim 14, wherein the processor is configured to:
extract a feature value expressing a feature from the prediction target peptide; and
generate a predicted value of biostability of the prediction target peptide by inputting the predictive feature vector including the docking profile and the feature value into the trained model.

### Claim 17

A trained model generation device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
compute a training docking profile that is a docking profile including a docking score of a plurality of training peptides by performing docking calculations between the respective training peptides and a blood plasma protein; and
generate a trained model for outputting a predicted value of biostability of a peptide from a feature vector including a docking profile obtained by docking calculation of a peptide by executing a machine learning algorithm on each of a plurality of the training peptides based on training data expressed by a training feature vector including the training docking profile paired with a correct value of biostability of the respective training peptide.

### Claim 18

A prediction device comprising:
a memory; and
a processor coupled to the memory, the processor being configured to:
extract a residue from a peptide that is a target for biostability prediction; and
predict biostability of the biostability prediction target peptide by, for each of a plurality of types of residue, reading a docking profile corresponding to the residue from the memory stored with a docking profile expressing a result of docking calculation between the residue and a blood plasma protein, and inputting a feature vector including a read docking profile of the prediction target residue into a trained model generated in advance by a machine learning algorithm.

### Claim 19

A prediction method comprising:
by a processor:
extracting a predictive feature vector expressing a feature from a peptide that is a target for biostability prediction;
adjusting a length of the extracted predictive feature vector to a prescribed length; and
generating a predicted value of biostability for the prediction target peptide by inputting the predictive feature vector adjusted in length into a trained model pre-trained to output a predicted value of peptide biostability from a feature vector expressing a feature of a peptide.

### Claim 20

A trained model generation method comprising:
by a processor:
extracting a training feature vector expressing a feature from each of a plurality of training peptides;
adjusting a length of each training feature vector for each of the plurality of extracted training peptides to a prescribed length, and
generating a trained model, for outputting a predicted value for peptide biostability of a peptide from a feature vector expressing a feature of a peptide by executing a machine learning algorithm based on training data that is the length-adjusted training feature vectors paired with correct values of biostability for the respective training peptides.

### Claim 21

A prediction method comprising:
by a processor:
extracting a predictive feature vector expressing a feature from a cyclic peptide that is a target for biostability prediction for instances in which each of a plurality of residues contained in the cyclic peptide is at a start point of a cyclic sequence; and
generating a predicted value of biostability of the prediction target cyclic peptide by inputting a plurality of the extracted predictive feature vectors into a trained model pre-trained to output a predicted value of biostability of a peptide from a feature vector expressing a feature of a cyclic peptide.

### Claim 22

A trained model generation method comprising:
by a processor:
extracting a training feature vector expressing a feature from each of a plurality of training cyclic peptides for instances in which each of a plurality of residues contained in the training cyclic peptide is at a start point of a cyclic sequence; and
generating a trained model for outputting a predicted value of biostability of a cyclic peptide from a feature vector expressing a feature of a cyclic peptide by executing a machine learning algorithm based on training data expressed by a plurality of training feature vectors extracted for each of a plurality of training cyclic peptides paired with correct values of biostability for the training cyclic peptides.

### Claim 23

A trained model generation method comprising:
by a processor:
extracting a first training feature vector expressing a feature from each of a plurality of training cyclic peptides;
generating a plurality of second training feature vectors for each of the extracted first training feature vectors by cyclically shifting elements of the first training feature vector, and generating training data expressed by the first training feature vector and the plurality of second training feature vectors paired with a correct value for biostability of the respective training cyclic peptide; and
generating a trained model for outputting a predicted value of biostability of a cyclic peptide from a feature vector expressing a feature of the cyclic peptide by executing a machine learning algorithm based on a plurality of the generated training data.

### Claim 24 (depends on 23)

A prediction method comprising:
by a processor:
extracting a predictive feature vector expressing a feature from a cyclic peptide that is a target for biostability prediction; and
generating a predicted value for biostability of the prediction target peptide by inputting the extracted predictive feature vector into a trained model generated by the trained model generation method of claim 23.

### Claim 25

A trained model generation method comprising:
by a processor:
generating a trained convolutional neural network model for outputting a predicted value of biostability of a cyclic peptide from a feature vector expressing a feature of a cyclic peptide by, based on training data expressed by a training feature vector expressing a feature extracted from each of a plurality of training cyclic peptides paired with a correct value of biostability for the plurality of respective training cyclic peptides, executing a machine learning algorithm using a convolutional neural network model including a both-end-adjacency layer in which elements at both ends of the training feature vector are placed adjacent to one another.

### Claim 26

A prediction method comprising:
by a processor:
extracting a predictive feature vector expressing a feature from a cyclic peptide that is a target for biostability prediction; and
generating a predicted value of biostability of the prediction target cyclic peptide by inputting the extracted predictive feature vector into a trained convolutional neural network model that is a trained convolutional neural network model including a both-end-adjacency layer in which elements at both ends of a feature vector expressing a feature of a cyclic peptide are placed adjacent to one another and that is configured to output a predicted value of biostability of a peptide from the feature vector.

### Claim 27

A prediction method comprising:
by a processor:
generating a plurality of conformations adoptable by a peptide that is a target for biostability prediction;
selecting a conformation to be subjected to docking calculation from the plurality of generated conformations based on a prescribed selection criteria; and
predicting the biostability of the prediction target peptide by performing docking calculation between a prediction target peptide corresponding to the selected conformation and a blood plasma protein.

### Claim 28

A prediction method comprising:
by a processor:
computing a predicted value of a first biostability expressing an biostability of a peptide that is a target for biostability prediction by performing docking calculation between the biostability prediction target peptide and a blood plasma protein;
generating a predicted value of a second biostability predicted value expressing biostability of the peptide by inputting a feature vector extracted from the prediction target peptide into a trained model generated in advance by a machine learning algorithm; and
computing biostability of the peptide by consolidating the generated first biostability predicted value with the second biostability predicted value.

### Claim 29

A prediction method comprising:
by a processor:
computing a docking profile including a docking score between a peptide that is a target for biostability prediction and a blood plasma protein by performing docking calculation between the peptide and the blood plasma protein; and
generating a predicted value of biostability for the prediction target peptide by inputting a predictive feature vector including the computed docking profile into a trained model generated in advance by a machine learning algorithm.

### Claim 30

A trained model generation method comprising:
by a processor:
computing a training docking profile that is a docking profile including a docking score of a training peptide by performing docking calculation between a plurality of training peptides and a blood plasma protein; and
generating a trained model for outputting a predicted value of biostability of a peptide from a feature vector including a docking profile obtained by performing docking calculation of the peptide by executing a machine learning algorithm based on training data expressed by a training feature vector including a computed training docking profile for each of a plurality of training peptides paired with a correct value of biostability for the respective training peptides.

### Claim 31

A prediction method comprising:
by a processor:
extracting a residue from a peptide that is a target for biostability prediction; and
predicting biostability of the prediction target peptide by reading a docking profile corresponding to the extracted residue from a storage section stored with docking profiles expressing results of docking calculations between a residue and a blood plasma protein for each of a plurality of types of residue, and by inputting a feature vector including the read docking profile of the prediction target residue into a trained model generated in advance by a machine learning algorithm.

### Claim 32 (depends on 19)

A non-transitory recording medium storing a prediction program executable by a computer to perform processing of the prediction method of claim 19.

### Claim 33 (depends on 20)

A non-transitory recording medium storing a trained model generation program executable by a computer to perform processing of the trained model generation method of claim 20.
