Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MetaEnzyme: Meta Pan-Enzyme Learning for Task-Adaptive Redesign

Jiangbin Zheng, Han Zhang, Qianqing Xu, An-Ping Zeng, Stan Z. Li

TL;DR

MetaEnzyme tackles the data-scarce landscape of enzyme design by unifying three low-resource tasks—FuncDesign, MutDesign, and SeqDesign—under a single, cross-modal framework. It pretrains a universal protein design network (UniProt-Net) with a geometry-aware structure-to-sequence encoder, energy-aware augmentation, and a language-informed context module, then adapts to tasks via few-shot meta-learning, zero-shot mutation scoring, and autoregressive sequence generation. The work demonstrates substantial improvements across function prediction, mutation effect ranking, and sequence generation, with strong in silico results and notable wet-lab validation that achieve high rank correlations in practical enzyme redesign scenarios. Collectively, MetaEnzyme advances generalizable, task-adaptive enzyme design by leveraging universal representations, geometry-aware learning, and domain-adaptive strategies to close the gap between computational predictions and experimental outcomes.

Abstract

Enzyme design plays a crucial role in both industrial production and biology. However, this field faces challenges due to the lack of comprehensive benchmarks and the complexity of enzyme design tasks, leading to a dearth of systematic research. Consequently, computational enzyme design is relatively overlooked within the broader protein domain and remains in its early stages. In this work, we address these challenges by introducing MetaEnzyme, a staged and unified enzyme design framework. We begin by employing a cross-modal structure-to-sequence transformation architecture, as the feature-driven starting point to obtain initial robust protein representation. Subsequently, we leverage domain adaptive techniques to generalize specific enzyme design tasks under low-resource conditions. MetaEnzyme focuses on three fundamental low-resource enzyme redesign tasks: functional design (FuncDesign), mutation design (MutDesign), and sequence generation design (SeqDesign). Through novel unified paradigm and enhanced representation capabilities, MetaEnzyme demonstrates adaptability to diverse enzyme design tasks, yielding outstanding results. Wet lab experiments further validate these findings, reinforcing the efficacy of the redesign process.

MetaEnzyme: Meta Pan-Enzyme Learning for Task-Adaptive Redesign

TL;DR

MetaEnzyme tackles the data-scarce landscape of enzyme design by unifying three low-resource tasks—FuncDesign, MutDesign, and SeqDesign—under a single, cross-modal framework. It pretrains a universal protein design network (UniProt-Net) with a geometry-aware structure-to-sequence encoder, energy-aware augmentation, and a language-informed context module, then adapts to tasks via few-shot meta-learning, zero-shot mutation scoring, and autoregressive sequence generation. The work demonstrates substantial improvements across function prediction, mutation effect ranking, and sequence generation, with strong in silico results and notable wet-lab validation that achieve high rank correlations in practical enzyme redesign scenarios. Collectively, MetaEnzyme advances generalizable, task-adaptive enzyme design by leveraging universal representations, geometry-aware learning, and domain-adaptive strategies to close the gap between computational predictions and experimental outcomes.

Abstract

Enzyme design plays a crucial role in both industrial production and biology. However, this field faces challenges due to the lack of comprehensive benchmarks and the complexity of enzyme design tasks, leading to a dearth of systematic research. Consequently, computational enzyme design is relatively overlooked within the broader protein domain and remains in its early stages. In this work, we address these challenges by introducing MetaEnzyme, a staged and unified enzyme design framework. We begin by employing a cross-modal structure-to-sequence transformation architecture, as the feature-driven starting point to obtain initial robust protein representation. Subsequently, we leverage domain adaptive techniques to generalize specific enzyme design tasks under low-resource conditions. MetaEnzyme focuses on three fundamental low-resource enzyme redesign tasks: functional design (FuncDesign), mutation design (MutDesign), and sequence generation design (SeqDesign). Through novel unified paradigm and enhanced representation capabilities, MetaEnzyme demonstrates adaptability to diverse enzyme design tasks, yielding outstanding results. Wet lab experiments further validate these findings, reinforcing the efficacy of the redesign process.
Paper Structure (34 sections, 15 equations, 9 figures, 4 tables, 1 algorithm)

This paper contains 34 sections, 15 equations, 9 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Universal Protein Design Network
  3. Problem Statement
  4. The Underlying Structure-to-Sequence Network
  5. Geometry-enhanced Structural Encoder
  6. Energy-motivated Data Augmentation
  7. Context Module Initialization
  8. Training Pipeline in Initial Phase
  9. Domain Adaption for Low-resource Enzyme Design Tasks
  10. Few-shot Learning for Functional Design (FuncDesign)
  11. Zero-shot Learning for Mutant Effect Prediction (MutDesign)
  12. Autoregressive Generalization for Sequence Generation (SeqDesign)
  13. In silico Experiments
  14. Setups
  15. Few-shot Learning for Function Prediction (FuncDesign)
  16. ...and 19 more sections

Figures (9)

  • Figure 1: The overall logical framework of MetaEnzyme and the successive phases for various enzyme functional tasks.
  • Figure 2: (a) The underlying UniProt-Net instantiation is based on a structure-to-sequence framework including a Geometry-invariant Structural Encoder (GeoStruc-Encoder), a Structure-Sequence Adapter (StrucSeq-Adapter), and a Self-attention Sequence Decoder (SaSeq-Decoder). (b) Elaborate feature modules within the GeoStruc-Encoder. (c) Structural data augmentation employing Riemann-Gaussian noise. (d) Initialization of the Context-Module (StrucSeq-Adapter & SaSeq-Decoder) in a self-supervised manner.
  • Figure 3: The proposed decoupled mutational effect scoring.
  • Figure 4: Comparing the folding prediction abilities of different models. P@k denotes the top-k precision.
  • Figure 5: Spearman’s rank correlation between predicted scores and experimental measurements on ProteinGym. Comparison among protein language models and inverse folding models.
  • ...and 4 more figures