# OPUS-BFactor: Predicting Protein B-Factor with Sequence and Structure Information

## TL;DR

OPUS-BFactor is a new tool that predicts protein flexibility using both sequence and structure information, outperforming existing methods.

## Contribution

A novel transformer-based method that integrates sequence and structural features for improved B-factor prediction.

## Key findings

- OPUS-BFactor outperforms existing B-factor prediction methods on recent test sets.
- The tool can operate using only sequence or structure information.
- It effectively combines pair features with sequence features using a transformer-based approach.

## Abstract

Protein B-factor, also known as the Debye–Waller temperature factor or atomic displacement parameter, measures the thermal fluctuation of an atom around its average position. It serves as a crucial indicator of protein flexibility and dynamics. However, accurately predicting the B-factor of Cα atoms remains challenging. In this work, we introduce OPUS-BFactor, a tool for predicting the normalized protein B-factor. OPUS-BFactor employs a transformer-based module to integrate sequence-level and pair-level features, encompassing structural attributes derived from the protein’s 3D structure and evolutionary profiles obtained from the protein language model ESM-2. Specifically, OPUS-BFactor treats pair features as a bias term, incorporating them into the attention matrix derived from the sequence-level features of each residue pair, thereby effectively merging pair features with sequence features. OPUS-BFactor operates in two modes, enabling predictions based solely on either the protein sequence or the 3D structure of the target protein. Evaluation on three test sets, including recently released targets from CAMEO and CASP15, demonstrated that OPUS-BFactor significantly outperformed other B-factor prediction methods. Therefore, OPUS-BFactor is a valuable tool for predicting protein properties related to the B-factor, such as flexibility, thermal stability, and regional activity.