Protein binder search
A specification of a binding target protein is received. A machine learning model is used to predict a plurality of candidates for a property of a selected amino acid position of a binder protein to bind to the binding target protein. For each selected property candidate of the plurality of property candidates, the selected property candidate is used as a model input to predict properties for one or more other amino acid positions into a corresponding candidate set of properties. The corresponding candidate sets are evaluated using a binding quality evaluation. Based on the evaluation, one of the plurality of property candidates is selected as a determined result property of the selected amino acid position. The determined result property is used as a model input to predict a plurality of candidates for a property of a different selected amino acid position included in the binder protein.
1. A method, comprising:
receiving a specification of a binding target protein;
receiving an identification of a known binding location of the binding target protein where a binder protein is to bind;
based at least in part on a property of the binding target protein at the known binding location, using a machine learning model to predict a plurality of candidates for a property of a selected amino acid position of the binder protein to bind to the binding target protein;
for each selected property candidate of the plurality of property candidates, using the selected property candidate as an input to the machine learning model to predict properties for one or more other amino acid positions of the binder protein into a corresponding candidate set of properties;
evaluating the corresponding candidate sets of properties for the plurality of property candidates using a binding quality evaluation;
based on the evaluation of the corresponding candidate set of properties, selecting one of the plurality of property candidates as a determined result property of the selected amino acid position of the binder protein in defining the binding target protein; and
iteratively refining at least a portion of the predicted properties for the one or more other amino acid positions of the binder protein including by discarding at least a portion of the predicted properties for the one or more other amino acid positions of the binder protein and feeding back the determined result property of the selected amino acid position as a subsequent feedback input to the machine learning model to predict a plurality of candidates for a property of a different selected amino acid position included in the binder protein.
2. The method of claim 1 , wherein the property of the selected amino acid position of the binder protein to bind to the binding target protein corresponds to an amino acid sequence token.
3. The method of claim 1 , wherein the property of the selected amino acid position of the binder protein to bind to the binding target protein corresponds to an amino acid structure token.
4. The method of claim 3 , wherein the structure token references neighboring amino acids of the binder protein and the binding target protein.
5. The method of claim 1 , further comprising selecting the selected amino acid position based on a search criterion.
6. The method of claim 5 , wherein the search criterion includes randomly selecting the selected amino acid position of the binder protein to bind to the binding target protein.
7. The method of claim 1 , further comprising decoding one or more structure tokens associated with the corresponding candidate set of properties.
8. The method of claim 7 , wherein using the binding quality evaluation includes analyzing the decoded one or more structure tokens.
9. The method of claim 1 , wherein using the binding quality evaluation includes determining distances between one or more atoms of the binder protein and one or more atoms of the binding target protein.
10. The method of claim 1 , wherein the known binding location corresponds to a specified epitope of the binding target protein.
11. The method of claim 1 , wherein using the binding quality evaluation includes accessing a computational oracle, a binding evaluation service, or a binding prediction model.
12. The method of claim 1 , further comprising:
creating an amino acid sequence token input sequence using at least one or more amino acid sequence tokens associated with the corresponding candidate set of properties;
providing the amino acid sequence token input sequence to the machine learning model to generate a proposed binder protein result;
predicting a structure result associated with the proposed binder protein result and the binding target protein; and
evaluating the structure result for a binding quality evaluation score.
13. The method of claim 1 , further comprising:
creating a structure token input sequence based on at least one or more structure tokens associated with the corresponding candidate set of properties;
providing the structure token input sequence to the machine learning model to generate a proposed binder protein result;
determining a structure result associated with the proposed binder protein result and the binding target protein; and
evaluating the structure result for a binding quality evaluation score.
14. The method of claim 1 , further comprising:
determining a first set of structure tokens representing the binding target protein, wherein the first set of structure tokens corresponds to a non-binding state of the binding target protein;
using the first set of structure tokens to determine a second set of structure tokens representing the binding target protein, wherein the second set of structure tokens correspond to a binding state of the binding target protein;
identifying one or more epitope structure tokens associated with an epitope of the binding target protein from the second set of structure tokens; and
including the identified one or more epitope structure tokens as an additional input to the machine learning model.
15. A system, comprising:
one or more processors configured to:
receive a specification of a binding target protein;
receive an identification of a known binding location of the binding target protein where a binder protein is to bind;
based at least in part on a property of the binding target protein at the known binding location, predict using a machine learning model a plurality of candidates for a property of a selected amino acid position of the binder protein to bind to the binding target protein;
for each selected property candidate of the plurality of property candidates, predict properties for one or more other amino acid positions of the binder protein into a corresponding candidate set of properties using the selected property candidate as an input to the machine learning model;
evaluate the corresponding candidate sets of properties for the plurality of property candidates using a binding quality evaluation;
based on the evaluation of the corresponding candidate set of properties, select one of the plurality of property candidates as a determined result property of the selected amino acid position of the binder protein in defining the binding target protein; and
iteratively refine at least a portion of the predicted properties for the one or more other amino acid positions of the binder protein including by being configured to discard at least a portion of the predicted properties for the one or more other amino acid positions of the binder protein and feed back the determined result property of the selected amino acid position as a subsequent feedback input to the machine learning model to predict a plurality of candidates for a property of a different selected amino acid position included in the binder protein; and
a memory coupled to at least one of the one or more processors and configured to provide instructions.
16. The system of claim 15 , wherein the property of the selected amino acid position of the binder protein to bind to the binding target protein corresponds to an amino acid sequence token.
17. The system of claim 15 , wherein the property of the selected amino acid position of the binder protein to bind to the binding target protein corresponds to an amino acid structure token.
18. The system of claim 17 , wherein the structure token references neighboring amino acids of the binder protein and the binding target protein.
19. The system of claim 15 , wherein using the binding quality evaluation includes determining distances between one or more atoms of the binder protein and one or more atoms of the binding target protein.
20. A computer program product, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:
receiving a specification of a binding target protein;
receiving an identification of a known binding location of the binding target protein where a binder protein is to bind;
based at least in part on a property of the binding target protein at the known binding location, using a machine learning model to predict a plurality of candidates for a property of a selected amino acid position of the binder protein to bind to the binding target protein;
for each selected property candidate of the plurality of property candidates, using the selected property candidate as an input to the machine learning model to predict properties for one or more other amino acid positions of the binder protein into a corresponding candidate set of properties;
evaluating the corresponding candidate sets of properties for the plurality of property candidates using a binding quality evaluation;
based on the evaluation of the corresponding candidate set of properties, selecting one of the plurality of property candidates as a determined result property of the selected amino acid position of the binder protein in defining the binding target protein; and
iteratively refining at least a portion of the predicted properties for the one or more other amino acid positions of the binder protein including by discarding at least a portion of the predicted properties for the one or more other amino acid positions of the binder protein and feeding back the determined result property of the selected amino acid position as a subsequent feedback input to the machine learning model to predict a plurality of candidates for a property of a different selected amino acid position included in the binder protein.