Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Pan-protein Design Learning Enables Task-adaptive Generalization for Low-resource Enzyme Design

Jiangbin Zheng, Ge Wang, Han Zhang, Stan Z. Li

TL;DR

This work introduces a novel CPD paradigm tailored for functional design tasks, particularly for enzymes a key protein class often lacking specific application efficiency, and presents CrossDesign, a domain-adaptive framework that leverages pretrained protein language models (PPLMs).

Abstract

Computational protein design (CPD) offers transformative potential for bioengineering, but current deep CPD models, focused on universal domains, struggle with function-specific designs. This work introduces a novel CPD paradigm tailored for functional design tasks, particularly for enzymes-a key protein class often lacking specific application efficiency. To address structural data scarcity, we present CrossDesign, a domain-adaptive framework that leverages pretrained protein language models (PPLMs). By aligning protein structures with sequences, CrossDesign transfers pretrained knowledge to structure models, overcoming the limitations of limited structural data. The framework combines autoregressive (AR) and non-autoregressive (NAR) states in its encoder-decoder architecture, applying it to enzyme datasets and pan-proteins. Experimental results highlight CrossDesign's superior performance and robustness, especially with out-of-domain enzymes. Additionally, the model excels in fitness prediction when tested on large-scale mutation data, showcasing its stability.

Pan-protein Design Learning Enables Task-adaptive Generalization for Low-resource Enzyme Design

TL;DR

This work introduces a novel CPD paradigm tailored for functional design tasks, particularly for enzymes a key protein class often lacking specific application efficiency, and presents CrossDesign, a domain-adaptive framework that leverages pretrained protein language models (PPLMs).

Abstract

Computational protein design (CPD) offers transformative potential for bioengineering, but current deep CPD models, focused on universal domains, struggle with function-specific designs. This work introduces a novel CPD paradigm tailored for functional design tasks, particularly for enzymes-a key protein class often lacking specific application efficiency. To address structural data scarcity, we present CrossDesign, a domain-adaptive framework that leverages pretrained protein language models (PPLMs). By aligning protein structures with sequences, CrossDesign transfers pretrained knowledge to structure models, overcoming the limitations of limited structural data. The framework combines autoregressive (AR) and non-autoregressive (NAR) states in its encoder-decoder architecture, applying it to enzyme datasets and pan-proteins. Experimental results highlight CrossDesign's superior performance and robustness, especially with out-of-domain enzymes. Additionally, the model excels in fitness prediction when tested on large-scale mutation data, showcasing its stability.

Paper Structure

This paper contains 18 sections, 8 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Main Structure-to-sequence Stream
  4. Auxiliary PPLM Supervised Stream
  5. Primary Generation Objective
  6. Cross-layer and Cross-modal Objective
  7. Inner-loop Sequence Reconstruction
  8. Sampling Regimes for Cross-domain Tasks
  9. AR Decoding for CPD
  10. NAR Decoding for Fitness
  11. Experiments
  12. Settings
  13. In-domain Evaluation on Generic Proteins.
  14. Out-of-Domain Enzyme Design
  15. Conditional CPD for PET Enzymes
  16. ...and 3 more sections

Figures (2)

  • Figure 1: (a). The proposed CrossDesign framework contains a Structure-to-Sequence Stream (Str2Seq Stream) and an auxiliary PPLM Supervised Stream (PPLM Stream). The PPLM stream is used to augment the Str2Seq stream using prior language knowledge, while the two streams share a decoder. CLC: Cross-Layer Consistency; InterMA: inter cross-modal alignment. Note that once the training is done, the framework performs the Str2Seq stream only for sampling. (b) Schematic diagram of Transformation-enhanced GVP (tGVP).
  • Figure 2: AAR of enzyme design. Orange for EnzPetDB; Green and blue for EnzFoldDB. EC/i(N) denotes the evaluation in the i-th fold of the number N.