Kelow, S. P. | bioRxiv Science

bioRxiv2025

Lab-in-the-loop therapeutic antibody design with deep learning

Therapeutic antibody design is a complex multi-property optimization problem with substantial promise for improvement with the application of machine-learning methods. Towards realizing that promise, we introduce "Lab-in-the-loop," a new approach that orchestrates state-of-the-art repertoire mining methods, generative machine learning models, multi-task property predictors, active learning ranking and selection, and in vitro experimentation in a semi-autonomous, iterative optimization loop. By automating the design of antibody variants, property prediction, ranking and selection of designs to assay in the lab, and ingestion of in vitro data, we enable an end-to-end approach to developing computationally-informed therapeutic antibody design pipelines. We apply lab-in-the-loop to eleven seed antibodies obtained via animal immunization with four clinically relevant antigen targets: EGFR, IL-6, HER2, and OSM. Over 1,800 unique antibody variants are tested throughout four rounds of iterative optimization identifying 3-100x better binding variants for all targets and 10/11 seeds, with the best binders exceeding 100 pM affinity, demonstrating a process by which end-to-end machine learning can be developed for therapeutic antibody development.

bioengineering↗

bioRxiv2020

Hiding in plain sight: structure and sequence analysis reveals the importance of the antibody DE loop for antibody-antigen binding.

Antibody variable domains contain "complementarity determining regions" (CDRs), the loops that form the antigen binding site. CDRs1-3 are recognized as the canonical CDRs. However, a fourth loop sits adjacent to CDR1 and CDR2 and joins the D and E strands on the antibody v-type fold. This "DE loop" is usually treated as a framework region, even though mutations in the loop affect the conformation of the CDRs and residues in the DE loop occasionally contact antigen. We analyzed the length, structure, and sequence features of all DE loops in the Protein Data Bank, as well as millions of sequences from HIV-1 infected and naive patients. We refer to the DE loop as H4 and L4 in the heavy and light chain respectively. Clustering the backbone conformations of the most common length of L4 (6 residues) reveals four conformations: two {kappa}-only clusters, one {lambda}-only cluster, and one mixed {kappa}/{lambda} cluster. The vast majority of H4 loops are length-8 and exist primarily in one conformation; a secondary conformation represents a small fraction of H4-8 structures. H4 sequence variability exceeds that of the antibody framework in naive human high-throughput sequences, and both L4 and H4 sequence variability from {lambda} and heavy germline sequences exceed that of germline framework regions. Finally, we identified dozens of structures in the PDB with insertions in the DE loop, all related to broadly neutralizing HIV-1 antibodies, as well as antibody sequences from high-throughput sequencing studies of HIV-infected individuals, illuminating a possible role in humoral immunity to HIV-1.

bioinformatics↗