Machine learning for functional protein design

Nature Biotechnology, Journal Year: 2024, Volume and Issue: 42(2), P. 216 - 228

Published: Feb. 1, 2024

Language: Английский

---

De novo protein design—From new structures to programmable functions

Cell, Journal Year: 2024, Volume and Issue: 187(3), P. 526 - 544

Published: Feb. 1, 2024

Methods from artificial intelligence (AI) trained on large datasets of sequences and structures can now "write" proteins with new shapes molecular functions de novo, without starting found in nature. In this Perspective, I will discuss the state field novo protein design at juncture physics-based modeling approaches AI. New folds higher-order assemblies be designed considerable experimental success rates, difficult problems requiring tunable control over conformations precise shape complementarity for recognition are coming into reach. Emerging incorporate engineering principles-tunability, controllability, modularity-into process beginning. Exciting frontiers lie deconstructing cellular and, conversely, constructing synthetic signaling ground up. As methods improve, many more challenges unsolved.

Language: Английский

---

Simulating 500 million years of evolution with a language model

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: July 2, 2024

Abstract More than three billion years of evolution have produced an image biology encoded into the space natural proteins. Here we show that language models trained on tokens generated by can act as evolutionary simulators to generate functional proteins are far away from known We present ESM3, a frontier multimodal generative model reasons over sequence, structure, and function ESM3 follow complex prompts combining its modalities is highly responsive biological alignment. prompted fluorescent with chain thought. Among generations synthesized, found bright protein at distance (58% identity) Similarly distant separated five hundred million evolution.

Language: Английский

---

Accelerating the integration of ChatGPT and other large‐scale AI models into biomedical research and healthcare

MedComm – Future Medicine, Journal Year: 2023, Volume and Issue: 2(2)

Published: May 17, 2023

Abstract Large‐scale artificial intelligence (AI) models such as ChatGPT have the potential to improve performance on many benchmarks and real‐world tasks. However, it is difficult develop maintain these because of their complexity resource requirements. As a result, they are still inaccessible healthcare industries clinicians. This situation might soon be changed advancements in graphics processing unit (GPU) programming parallel computing. More importantly, leveraging existing large‐scale AIs GPT‐4 Med‐PaLM integrating them into multiagent (e.g., Visual‐ChatGPT) will facilitate implementations. review aims raise awareness applications healthcare. We provide general overview several advanced AI models, including language vision‐language graph learning language‐conditioned multimodal embodied models. discuss medical addition challenges future directions. Importantly, we stress need align with human values goals, using reinforcement from feedback, ensure that accurate personalized insights support decision‐making outcomes.

Language: Английский

---

Opportunities and Challenges for Machine Learning-Assisted Enzyme Engineering

ACS Central Science, Journal Year: 2024, Volume and Issue: 10(2), P. 226 - 241

Published: Feb. 5, 2024

Enzymes can be engineered at the level of their amino acid sequences to optimize key properties such as expression, stability, substrate range, and catalytic efficiency-or even unlock new activities not found in nature. Because search space possible proteins is vast, enzyme engineering usually involves discovering an starting point that has some desired activity followed by directed evolution improve its "fitness" for a application. Recently, machine learning (ML) emerged powerful tool complement this empirical process. ML models contribute (1) discovery functional annotation known protein or generating novel with functions (2) navigating fitness landscapes optimization mappings between associated values. In Outlook, we explain how complements discuss future potential improved outcomes.

Language: Английский

---

Bilingual Language Model for Protein Sequence and Structure

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: July 25, 2023

Abstract Adapting large language models (LLMs) to protein sequences spawned the development of powerful (pLMs). Concurrently, AlphaFold2 broke through in structure prediction. Now we can systematically and comprehensively explore dual nature proteins that act exist as three-dimensional (3D) machines evolve linear strings one-dimensional (1D) sequences. Here, leverage pLMs simultaneously model both modalities by combining 1D with 3D a single model. We encode structures token using 3Di-alphabet introduced 3D-alignment method Foldseek . This new foundation pLM extracts features patterns resulting “structure-sequence” representation. Toward this end, built non-redundant dataset from AlphaFoldDB fine-tuned an existing (ProtT5) translate between 3Di amino acid As proof-of-concept for our novel approach, dubbed Protein structure-sequence T5 ( ProstT5 ), showed improved performance subsequent prediction tasks, “inverse folding”, namely generation adopting given structural scaffold (“fold”). Our work showcased potential tap into information-rich revolution fueled AlphaFold2. paves way develop tools integrating vast resource predictions, opens research avenues post-AlphaFold2 era. is freely available all at https://github.com/mheinzinger/ProstT5

Language: Английский

---

Protein generation with evolutionary diffusion: sequence is all you need

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Sept. 12, 2023

Abstract Deep generative models are increasingly powerful tools for the in silico design of novel proteins. Recently, a family called diffusion has demonstrated ability to generate biologically plausible proteins that dissimilar any actual seen nature, enabling unprecedented capability and control de novo protein design. However, current state-of-the-art structures, which limits scope their training data restricts generations small biased subset space. Here, we introduce general-purpose framework, EvoDiff, combines evolutionary-scale with distinct conditioning capabilities controllable generation sequence EvoDiff generates high-fidelity, diverse, structurally-plausible cover natural functional We show experimentally express, fold, exhibit expected secondary structure elements. Critically, can inaccessible structure-based models, such as those disordered regions, while maintaining scaffolds structural motifs. validate universality our sequence-based formulation by characterizing intrinsically-disordered mitochondrial targeting signals, metal-binding proteins, binders designed using EvoDiff. envision will expand engineering beyond structure-function paradigm toward programmable, sequence-first

Language: Английский

---

xTrimoPGLM: Unified 100B-Scale Pre-trained Transformer for Deciphering the Language of Protein

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: July 6, 2023

Protein language models have shown remarkable success in learning biological information from protein sequences. However, most existing are limited by either autoencoding or autoregressive pre-training objectives, which makes them struggle to handle understanding and generation tasks concurrently. We propose a unified model, xTrimoPGLM, address these two types of simultaneously through an innovative framework. Our key technical contribution is exploration the compatibility potential for joint optimization has led strategy training xTrimoPGLM at unprecedented scale 100 billion parameters 1 trillion tokens. extensive experiments reveal that 1) significantly outperforms other advanced baselines 18 benchmarks across four categories. The model also facilitates atomic-resolution view structures, leading 3D structural prediction surpasses model-based tools. 2) not only can generate de novo sequences following principles natural ones, but perform programmable after supervised fine-tuning (SFT) on curated These results highlight substantial capability versatility generating sequences, contributing evolving landscape foundation science.

Language: Английский

---

Opportunities and challenges in design and optimization of protein function

Nature Reviews Molecular Cell Biology, Journal Year: 2024, Volume and Issue: 25(8), P. 639 - 653

Published: April 2, 2024

Language: Английский

---

Simulating 500 million years of evolution with a language model

Science, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 16, 2025

More than three billion years of evolution have produced an image biology encoded into the space natural proteins. Here we show that language models trained at scale on evolutionary data can generate functional proteins are far away from known We present ESM3, a frontier multimodal generative model reasons over sequence, structure, and function ESM3 follow complex prompts combining its modalities is highly responsive to alignment improve fidelity. prompted fluorescent Among generations synthesized, found bright protein distance (58% sequence identity) proteins, which estimate equivalent simulating five hundred million evolution.

Language: Английский

---