PROTEIN REFINEMENT AND JOINT OPTIMIZATION
Information is received for at least a portion of a first track included in a plurality of tracks for one or more generative protein language models. Based at least in part on the received information, at least one of the one or more generative protein language models is used to predict at least a portion of a second track of the plurality of tracks. Values of the plurality of tracks are iteratively refined including by iteratively alternating between different selected tracks of the plurality of tracks as input conditions to at least one of the one or more generative protein language models to update values of at least one of the plurality of tracks.
1 . A method, comprising:
receiving information for at least a portion of a first track included in a plurality of tracks for one or more generative protein language models;
based at least in part on the received information, using at least one of the one or more generative protein language models to predict at least a portion of a second track of the plurality of tracks; and
iteratively refining values of the plurality of tracks including by iteratively alternating between different selected tracks of the plurality of tracks as input conditions to at least one of the one or more generative protein language models to update values of at least one of the plurality of tracks.
2 . The method of claim 1 , wherein the plurality of tracks includes at least a protein sequence track and a protein structure track.
3 . The method of claim 1 , wherein the plurality of tracks includes a function track, a secondary structure track, or a protein solvent accessible surface area track.
4 . The method of claim 2 , wherein predicting at least the portion of the second track includes generating a structural representation of a candidate protein design based on the protein sequence track.
5 . The method of claim 2 , wherein predicting at least the portion of the second track includes generating a sequence representation of a candidate protein design based on the protein structure track.
6 . The method of claim 1 , wherein iteratively refining values of the plurality of tracks includes alternating conditions for the first track and the second track.
7 . The method of claim 1 , wherein at least one of the one or more generative protein language models is a multi-track model configured to receive inputs for multiple biological tracks and generate outputs for at least one biological track.
8 . The method of claim 1 , wherein at least one of the one or more generative protein language models is a protein structure prediction model based on a diffusion architecture or a transformer-based attention mechanism.
9 . The method of claim 1 , further comprising evaluating one or more generated candidate designs using a biological evaluation metric.
10 . The method of claim 9 , wherein the biological evaluation metric corresponds to a predicted stability score, a structure-sequence compatibility score, a folding confidence score, a binding affinity score, or an expression likelihood score.
11 . The method of claim 9 , further comprising selecting one or more of the one or more generated candidate designs for continued refinement based on the biological evaluation metric.
12 . The method of claim 1 , further comprising terminating the iterative refining of values of the plurality of tracks when one or more stopping criteria are satisfied.
13 . The method of claim 12 , wherein at least one of the one or more stopping criteria is based on a convergence of predictions, a satisfaction of threshold metrics, or a maximum number of performed iterations.
14 . A system, comprising:
one or more processors configured to:
receive information for at least a portion of a first track included in a plurality of tracks for one or more generative protein language models;
based at least in part on the received information, use at least one of the one or more generative protein language models to predict at least a portion of a second track of the plurality of tracks; and
iteratively refine values of the plurality of tracks including by iteratively alternating between different selected tracks of the plurality of tracks as input conditions to at least one of the one or more generative protein language models to update values of at least one of the plurality of tracks; and
a memory coupled to the one or more processors, wherein the memory is configured to provide the one or more processors with instructions.
15 . The system of claim 14 , wherein the plurality of tracks includes at least a protein sequence track and a protein structure track.
16 . The system of claim 15 , wherein predicting at least the portion of the second track includes generating a structural representation of a candidate protein design based on the protein sequence track.
17 . The system of claim 15 , wherein predicting at least the portion of the second track includes generating a sequence representation of a candidate protein design based on the protein structure track.
18 . The system of claim 14 , wherein iteratively refining values of the plurality of tracks includes alternating conditions for the first track and the second track.
19 . The system of claim 14 , wherein at least one of the one or more generative protein language models is a multi-track model configured to receive inputs for multiple biological tracks and generate outputs for at least one biological track.
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 information for at least a portion of a first track included in a plurality of tracks for one or more generative protein language models;
based at least in part on the received information, using at least one of the one or more generative protein language models to predict at least a portion of a second track of the plurality of tracks; and
iteratively refining values of the plurality of tracks including by iteratively alternating between different selected tracks of the plurality of tracks as input conditions to at least one of the one or more generative protein language models to update values of at least one of the plurality of tracks.