BioPhi: A platform for antibody design, humanization and humanness evaluation based on natural antibody repertoires and deep learning

Abstract

Despite recent advances in transgenic animal models and display technologies, humanization of mouse sequences remains the primary route for therapeutic antibody development. Traditionally, humanization is manual, laborious, and requires expert knowledge. Although automation efforts are advancing, existing methods are either demonstrated on a small scale or are entirely proprietary. To predict the immunogenicity risk, the human-likeness of sequences can be evaluated using existing humanness scores, but these lack diversity, granularity or interpretability. Meanwhile, immune repertoire sequencing has generated rich antibody libraries such as the Observed Antibody Space (OAS) that offer augmented diversity not yet exploited for antibody engineering. Here we present BioPhi, an open-source platform featuring novel methods for humanization (Sapiens) and humanness evaluation (OASis). Sapiens is a deep learning humanization method trained on the OAS database using language modeling. Based on an in silico humanization benchmark of 177 antibodies, Sapiens produced sequences at scale while achieving results comparable to that of human experts. OASis is a granular, interpretable and diverse humanness score based on 9-mer peptide search in the OAS. OASis separated human and non-human sequences with high accuracy, and correlated with clinical immunogenicity. Together, BioPhi offers an antibody design interface with automated methods that capture the richness of natural antibody repertoires to produce therapeutics with desired properties and accelerate antibody discovery campaigns. BioPhi is accessible at https://biophi.dichlab.org and https://github.com/Merck/BioPhi.

Progress and challenges for the machine learning-based design of fit-for-purpose monoclonal antibodies

“… 251 Most recently, Prihoda et al 57 , 252 devised an in silico platform BioPhi that offers three complementary tools: 1) OASis, short for Observed Antibody Space (OAS) identity search, is an interpretable humanness scoring system based on an exact 9-mer peptide search within the OAS database, capable of accurately distinguishing human and non-human sequences with clinical immunogenicity correlation; 2) Sapiens, is an ML-based humanization method trained on the OAS human database using language modeling to recognize and substitute non-human sequences with human native equivalents in FR regions to improve sequence humanness (the OASis score); and 3) an interactive interface, to incorporate AA substitutions in the sequence and visualization. 57 , 252 In their study, Prihoda et al 57 , 252 compared the humanization performance of Sapiens on 152 precursor sequences of humanized mAbs against Hu-mAb (computational) and mutation-based humanization (experimental). They reported that Sapiens achieved higher humanness improvement than Hu-mAb and comparable results to experimental methods, suggesting AA substitutions that were experimentally validated to be advantageous for sequence humanization, while maintaining mAb specificity and binding affinity.…”

Deciphering the language of antibodies using self-supervised learning

“…AntiBERTa is pre-trained via MLM, which has been used elsewhere (26,29,35). Briefly, 15% of amino acids are chosen for perturbation.…”

“…AntiBERTa is pre-trained using a self-supervised MLM task, like other transformer-based protein LMs (26,29,35). Briefly, 15% amino acids within the input BCR sequence are randomly perturbed, and the model determines the correct amino acid in place of these masked positions.…”

“…A rapid, accurate, paratope prediction method can shed light on binding properties, and is valuable for therapeutic antibody engineering (42,43,49). While we focus on paratope prediction here, the AntiBERTa model can potentially be fine-tuned for other tasks such as antibody structure prediction and humanisation (13,35).…”

“…To date, two LMs have been deployed for BCRs and antibodies: DeepAb (13) and Sapiens (35). DeepAb is a bi-directional long short-term memory (LSTM) network that is pre-trained on 100k paired BCR sequences from the Observed Antibody Space (36,37).…”