Protocol for tri-modal pretraining and downstream fine-tuning of protein representations using ProteinAligner.

Abstract

ProteinAligner integrates protein sequence, three-dimensional structure, and functional annotations in a shared embedding space. Here, we present a protocol for tri-modal pretraining and downstream fine-tuning of protein representations using ProteinAligner. We describe steps for preparing both paired sequence-structure and sequence-text datasets from public protein resources and pretraining ProteinAligner with distributed multimodal learning. We then detail procedures for fine-tuning downstream tasks and evaluating performance. For complete details on the use and execution of this protocol, please refer to Zhang et al.¹.