Biology subjects

Chuai, G.

Publications and source records attributed to Chuai, G..

6 recordsLinked to original sources

bioRxiv2026

Towards building a World Model to simulate perturbation-induced cellular dynamics by AlphaCell

Predicting cellular responses to perturbations is crucial for therapeutic discovery, yet experimental screening is severely constrained by the combinatorial vastness of biological space. While computational simulations offer a scalable alternative, current models are limited by the incomplete latent representation----mainly relying on highly variable genes in feature representation; the poor genome-wide reconstruction fidelity; and the ungeneralizable dynamic laws across diverse contexts. Consequently, they fail to mechanistically transfer learned dynamics to unseen cellular contexts. To address these systemic flaws, we introduce AlphaCell, a generative Virtual Cell World Model that unifies genome-wise representation with continuous state transition modeling. AlphaCell achieves three synergistic innovations: (1) Latent Manifold Rectification, processing the full protein-coding transcriptome to construct a differentiable Virtual Cell Space, effectively filtering noise while preserving intrinsic cellular topology; (2) Biological Reality Reconstruction, utilizing a massive, knowledge-rich decoder to translate abstract latent states back into high-fidelity, genome-wide expression profiles; and (3) Universal State Transition, applying Optimal Transport Conditional Flow Matching to model perturbations as continuous, deterministic vector fields. By abstracting perturbation mechanisms into generalized dynamic laws, AlphaCell makes robust prediction of perturbation responses in a compositional generalization scenario and enables zero-shot prediction of cellular dynamics in entirely unseen cellular contexts, providing a foundational engine for cellular-context-generalizable perturbation prediction and perturbation-induced cellular dynamics simulation.

bioinformatics↗

bioRxiv2024

Discovering CRISPR-Cas system with self-processing pre-crRNA capability by foundation models

The discovery and functional annotation of CRISPR-Cas systems laid the groundwork for the development of novel CRISPR-based gene editing tools. Traditional similarity- search-based Cas discovery strategies, which rely heavily on local sequence alignment and reference Cas homologs, may overlook a significant number of remote homologs with limited sequence similarity; and it can not be applied directly for functional recognition. With the rapid development of protein large language models (LLMs), protein foundation models are expected to help model Cas systems with limited Cas homologs without extensive task-specific training data; however, the full potential of these models for Cas discovery and functional annotation has yet to be determined. To this end, we present a novel, effective and unified AI framework, CHOOSER (Cas HOmlog Observing and SElf-processing scReening), for alignment-free discovery of novel CRISPR-Cas systems with self-processing precursor CRISPR RNA (pre-crRNA) capability utilizing protein foundation models. CHOOSER successfully retrieved 11 novel homologs of Cas{lambda}, the majority of which are predicted to be able to self-process pre-crRNA, nearly doubling the current catalog. One of the candidates, EphcCas{lambda}, was subsequently experimentally validated for its ability to self-process pre-crRNA, target DNA cleavage, and trans-cleavage and was shown to be a promising candidate for use as a CRISPR-Cas-based pathogen detection system. Overall, our study provides an unprecedented perspective and methodology for discovering novel CRISPR-Cas systems with specific functions using foundation models, underscoring the potential for transforming newly identified Cas homologs into genetic editing tools.

bioinformatics↗

bioRxiv2024

Toward subtask decomposition-based learning and benchmarking for genetic perturbation outcome prediction and beyond

Deciphering cellular responses to genetic perturbations is fundamental for a wide array of biomedical applications, ranging from uncovering gene roles and interactions to unraveling effective therapeutics. Accurately predicting the transcriptional outcomes of genetic perturbations is indispensable for optimizing experimental perturbations and deciphering cellular response mechanisms; however, three scenarios present principal challenges, i.e., predicting single genetic perturbation outcomes, predicting multiple genetic perturbation outcomes and predicting genetic outcomes across cell lines. In this study, we introduce SubTAsk decomposition Modeling for genetic Perturbation prediction (STAMP), a conceptually novel computational strategy for genetic perturbation outcome prediction and downstream applications. STAMP innovatively formulates genetic perturbation prediction as a subtask decomposition (STD) problem by resolving three progressive subtasks in a divide-and-conquer manner, i.e., identifying differentially expressed gene (DEG) postperturbations, determining the regulatory directions of DEGs and finally estimating the magnitudes of gene expression changes. In addition to facilitating perturbation prediction, STAMP also serves as a robust and generalizable benchmark guide for evaluating various genetic perturbation prediction models. As a result, STAMP exhibits a substantial improvement in terms of its genetic perturbation prediction ability over the existing approaches on three subtasks and beyond, including revealing the ability to identify key regulatory genes and pathways on small samples and to reveal precise genetic interactions. Overall, STAMP serves as a fundamentally novel and effective prediction and generalizable benchmarking strategy that can facilitate genetic perturbation prediction, guide the design of perturbation experiments, and broaden the understanding of perturbation mechanisms.

bioinformatics↗

bioRxiv2020

X-MOL: large-scale pre-training for molecular understanding and diverse molecular analysis

In silico modelling and analysis of small molecules substantially accelerates the process of drug development. Representing and understanding molecules is the fundamental step for various in silico molecular analysis tasks. Traditionally, these molecular analysis tasks have been investigated individually and separately. In this study, we presented X-MOL, which applies large-scale pre-training technology on 1.1 billion molecules for molecular understanding and representation, and then, carefully designed fine-tuning was performed to accommodate diverse downstream molecular analysis tasks, including molecular property prediction, chemical reaction analysis, drug-drug interaction prediction, de novo generation of molecules and molecule optimization. As a result, X-MOL was proven to achieve state-of-the-art results on all these molecular analysis tasks with good model interpretation ability. Collectively, taking advantage of super large-scale pre-training data and super-computing power, our study practically demonstrated the utility of the idea of "mass makes miracles" in molecular representation learning and downstream in silico molecular analysis, indicating the great potential of using large-scale unlabelled data with carefully designed pre-training and fine-tuning strategies to unify existing molecular analysis tasks and substantially enhance the performance of each task.

bioinformatics↗

bioRxiv2020

FL-QSAR: a federated learning based QSAR prototype for collaborative drug discovery

MotivationQuantitative structure-activity relationship (QSAR) analysis is commonly used in drug discovery. Collaborations among pharmaceutical institutions can lead to a better performance in QSAR prediction, however, intellectual property and related financial interests remain substantially hindering inter-institutional collaborations in QSAR modeling for drug discovery. ResultsFor the first time, we verified the feasibility of applying the horizontal federated learning (HFL), which is a recently developed collaborative and privacy-preserving learning framework to perform QSAR analysis. A prototype platform of federated-learning-based QSAR modeling for collaborative drug discovery, i.e, FL-QSAR, is presented accordingly. We first compared the HFL framework with a classic privacy-preserving computation framework, i.e., secure multiparty computation (MPC) to indicate its difference from various perspective. Then we compared FL-QSAR with the public collaboration in terms of QSAR modeling. Our extensive experiments demonstrated that (1) collaboration by FL-QSAR outperforms a single client using only its private data, and (2) collaboration by FL-QSAR achieves almost the same performance as that of collaboration via cleartext learning algorithms using all shared information. Taking together, our results indicate that FL-QSAR under the HFL framework provides an efficient solution to break the barriers between pharmaceutical institutions in QSAR modeling, therefore promote the development of collaborative and privacy-preserving drug discovery with extendable ability to other privacy-related biomedical areas. Availability and implementationThe source codes of the federated learning simulation and FL-QSAR are available on the GitHub: https://github.com/bm2-lab/FL-QSAR

bioinformatics↗

bioRxiv2017

Unexpected CRISPR off-target mutation pattern in vivo are not typicallygermline-like

To the EditorSchaefer et al.1 (referred to as Study_1) recently presented the provocative conclusion that CRISPR-Cas9 nuclease can induce many unexpected off-target mutations across the genome that arise from the sites with poor homology to the gRNA. As Wilson et al.2 pointed out, however, the selection of a co-housed mouse as the control is insufficient to attribute the observed mutation differences between the CRISPR-treated mice and control mice. Therefore, the causes of these mutations need to be further investigated. In 2015, Iyer et al.3 (referred to as Study_2) used Cas9 and a pair of sgRNAs to mutate the Ar gene in vivo and off-target mutations were investigated by comparison the control mice and the offspring of the modified mice. After analyzing the whole genome sequencing (WGS) of the offspring and the control mice, they claimed that off-target mutations are rare from CRISPR-Cas9 engineering. Notably, their study only focused on indel off-target mutations. We re-analyzed the WGS data of these two studies and detected both single nucleotide variants (SNVs) and indel mutations.

genetics↗