# US 12,603,153 B2 · De novo generation of high diversity proteins in silico with selective affinity and cross-reactivity minimization > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,603,153 B2 - Document ID: US12603153B2 - Status: Granted - Kind: B2 - Application number: 17/396,197 - Filed: 2021-08-06 - Published: 2022-03-24 - Granted: 2026-04-14 - Earliest claimed: 2020-08-12 - Assignees: Thaumachron LLC - Applicants: Thaumachron LLC - Inventors: Andrew Simpson - Web page: https://patentgrep.com/patent/US12603153B2 ## 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 - G16H 20/10: relating to drugs or medications, e.g. for ensuring correct administration to patients - G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA ## Abstract This disclosure relates to a platform for de novo in silico generation of high diversity protein complexes, namely antibodies and TCRs, with selective affinity and cross-reactivity minimization given the antigen complex's sequences and additional conditional information. ## Claims ### Claim 1 A system for de novo generation of protein complexes in silico, the system comprising: a system processor connected to an input/output interface and a memory via a communications network, wherein said processor is programmed with a generative model, wherein said generative model is programmed to generate a plurality of high diversity protein complex outputs de novo using a first set of input information, the first set of input information comprising one or more antigen complex conditional information input value, wherein said protein complex outputs each have selective affinity and cross-reactivity minimization based on said first set of input information, wherein said generative model comprises one or more generative adversarial network (GAN), and wherein said generative model is further trained with one or more neural network module with one of the group consisting of: protein data, immune epitope data, antibody affinity data, TCR affinity data, off-target binding data, and combinations thereof; a protein folding module configured to assist said generative model to produce three-dimensional (3D) structural data for said protein complex outputs; and a discriminator module configured to compare said protein complex outputs with real protein complex data for continuous training of said generative model. ### Claim 2 (depends on 1) The system of claim 1, wherein said protein complex outputs are selected from the group consisting of: antibody outputs, T-cell receptor (TCR) outputs, and combinations thereof. ### Claim 3 (depends on 1) The system of claim 1, wherein said input information is selected from the group consisting of: an antigen complex sequence, species information, isotype information, subtype information, allotype information, affinity information, cross-reactivity conditions, and combinations thereof. ### Claim 4 (depends on 1) The system of claim 1, wherein said protein complex outputs comprise protein complex distance matrices, angle information, and surface mesh representations. ### Claim 5 (depends on 1) The system of claim 1, wherein said protein complex outputs comprise protein complex affinity information, epitope information, and paratope information. ### Claim 6 (depends on 1) The system of claim 1, further comprising: an affinity module configured to predict affinity between said protein complex outputs and an antigen to be targeted to update generated affinity output to match expected affinity and minimize cross-reactivity of said protein complex outputs. ### Claim 7 (depends on 1) The system of claim 1, wherein: said system is further configured to send said protein complex outputs to an oligonucleotide synthesizer to begin creation of proteins or cell lines according to said protein complex outputs. ### Claim 8 (depends on 1) The system of claim 1, wherein said input/output interface is accessible to a user via a web-based platform. ### Claim 9 A method of generating high diversity protein complexes in silico, the method comprising: providing a system processor connected to an input/output interface and a memory via a communications network, wherein said processor is programmed with a generative model, and wherein said generative model is trained using one or more neural network module with protein data, immune epitope data, antibody affinity data, TCR affinity data, and off-target binding data and programmed for generation of high diversity protein complexes de novo to target an antigen of using a first set of input information, wherein the first set input information comprises one or more antigen complex conditional information input value; providing a protein folding module configured to assist said generative model to produce three-dimensional (3D) structural data for said protein complex outputs; providing a discriminator module configured to compare said protein complex outputs with real protein complex data for continuous training of said generative model; inputting the first set of input information into said system processor; and generating a plurality of high diversity protein complex outputs de novo with selective affinity and cross-reactivity minimization based on said first set of input information. ### Claim 10 (depends on 9) The method of claim 9, wherein said first set of input information is selected from the group consisting of: an antigen complex sequence, species information, isotype information, subtype information, allotype information, affinity information, cross-reactivity conditions, and combinations thereof. ### Claim 11 (depends on 9) The method of claim 9, wherein said protein complex outputs comprise protein complex distance matrices, angle information, surface mesh representations, affinity, epitope, and paratope information. ### Claim 12 (depends on 9) The method of claim 9, further comprising: sending said protein complex outputs directly to an oligonucleotide synthesizer: said oligonucleotide synthesizer creating proteins or cell lines according to said protein complex outputs; and testing for safety of said protein complex outputs in vitro. ### Claim 13 (depends on 9) The method of claim 9, further comprising: performing an exhaustive search for self-reactivity of said protein complex outputs. ### Claim 14 (depends on 9) The method of claim 9, further comprising: performing an exhaustive search for cross-reactivity with all proteins for a particular species. ### Claim 15 (depends on 9) The method of claim 9, wherein said first set of input information comprises tumor specific antigen complex conditional information unique to an individual. ### Claim 16 (depends on 9) The method of claim 9, further comprising: further training said generative model with said protein complex outputs to further minimize cross-reactivity of additional protein complex outputs. ### Claim 17 A system for de novo generation of protein complexes in silico, the system comprising: a system processor connected to an input/output interface and a memory via a communications network: wherein said processor is programmed with a generative model: wherein said generative model is programmed to generate a plurality of high diversity protein complex outputs de novo using a first set of input information, the first set of input information comprising one or more antigen complex conditional information input value; wherein said input information is selected from the group consisting of: an antigen complex sequence, species information, isotype information, subtype information, allotype information, affinity information, cross-reactivity conditions, and combinations thereof; wherein said generative model comprises one or more generative adversarial network (GAN); wherein said generative model is further trained with one or more neural network module with one of the group consisting of: protein data, immune epitope data, antibody affinity data, TCR affinity data, off-target binding data, and combinations thereof; a protein folding module configured to assist said generative model to produce three-dimensional (3D) structural data for said protein complex outputs; an affinity module configured to predict affinity between said protein complex outputs and an antigen to be targeted to update generated affinity output to match expected affinity and minimize cross-reactivity of said protein complex outputs; wherein said protein complex outputs are selected from the group consisting of: antibody outputs, T-cell receptor (TCR) outputs, and combinations thereof; wherein said protein complex outputs each have selective affinity and cross-reactivity minimization based on said first set of input information; and wherein said protein complex outputs comprise protein complex affinity information, epitope information, and paratope information.