Machine learning method for protein modelling to design engineered peptides
Inventors
Greving, Matthew P. • TAGUCHI, Alexander T. • Hauser, Kevin E.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11545238-B2
Publication Date
2023-01-03
Expiration Date
Abstract
Provided herein are methods for design of engineered polypeptides that recapitulate molecular structure features of a predetermined portion of a reference protein structure, e.g., an antibody epitope or a protein binding site. A Machine Learning (ML) model is trained by labeling blueprint records generated from a reference target structure with scores calculated based on computational protein modeling of polypeptide structures generated by the blueprint records. The method may include training an ML model based on a first set of blueprint records, or representations thereof, and a first set of scores, each blueprint record from the first set of blueprint records associated with each score from the first set of scores. After the training, the machine learning model may be executed to generate a second set of blueprint records. A set of engineered polypeptides are then generated based on the second set of blueprint records.
Core Innovation
The invention designs engineered polypeptides using a machine learning model by receiving a representation of a reference target structure for a reference target. A predetermined portion of the reference target structure is used to generate a training set of blueprint records, where each blueprint record comprises target residue positions and scaffold residue positions.
Each blueprint record in the training set is labeled with a score by performing computational protein modeling on the blueprint record to generate a polypeptide structure, calculating a score for the polypeptide structure, and associating the score with that blueprint record. The machine learning model is trained based on the labeled training set and is applied to a set of desired scores to generate an output set of blueprint records with the desired scores.
The invention supports implementing the method on a non-transitory processor-readable medium that stores code representing instructions for executing the same blueprint-record workflow. Additional refinements described include constraining blueprint records via selection or ordering of target residue positions, using computational protein modeling based on de novo design without template matching, and expanding score composition to include an energy term and a structure-constraint matching term extracted from a reference target structure representation.
Claims Coverage
The document includes two independent claims, a method claim and a non-transitory processor-readable medium claim. Both claims share the same core inventive workflow: generate blueprint records from a portion of a reference target structure, computationally label them with scores, train a machine learning model, and apply the trained model to desired scores to generate output blueprint records.
Blueprint records from a portion of a reference target structure
Generating a training set of blueprint records from a predetermined portion of the reference target structure, each blueprint record comprising target residue positions and scaffold residue positions, with each target residue position corresponding to one target residue from a plurality of target residues.
Computational protein modeling for score-labeled training
Labeling each blueprint record of the training set with a score by performing computational protein modeling on the blueprint record to generate a polypeptide structure, calculating a score for the polypeptide structure, and associating the score with that blueprint record.
Machine learning model trained from labeled blueprint records
Training a machine learning model based on the labeled training set.
Applying the trained model to desired scores to generate output blueprints
Applying the trained machine learning model to a set of desired scores to generate an output set of blueprint records with the desired scores.
Non-transitory processor-readable medium implementing the same workflow
A non-transitory processor-readable medium storing code representing instructions executed by a processor to receive a representation of a reference target structure, generate training blueprint records, label them with computational modeling scores, train a machine learning model, and apply the trained model to desired scores to generate output blueprint records with the desired scores.
Across the two independent claims, the inventive coverage centers on blueprint records derived from a portion of a reference target structure, computational protein modeling to create score-labeled training data, supervised training of a machine learning model, and application of the trained model to desired scores to output blueprint records.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?