arxiv_ml 40% Match Research Paper Computational biologists,Bioinformaticians,Drug discovery scientists,Machine learning researchers in biology 3 weeks ago

scPPDM: A Diffusion Model for Single-Cell Drug-Response Prediction

computer-vision › diffusion-models

📄 Abstract

Abstract: This paper introduces the Single-Cell Perturbation Prediction Diffusion Model (scPPDM), the first diffusion-based framework for single-cell drug-response prediction from scRNA-seq data. scPPDM couples two condition channels, pre-perturbation state and drug with dose, in a unified latent space via non-concatenative GD-Attn. During inference, factorized classifier-free guidance exposes two interpretable controls for state preservation and drug-response strength and maps dose to guidance magnitude for tunable intensity. Evaluated on the Tahoe-100M benchmark under two stringent regimes, unseen covariate combinations (UC) and unseen drugs (UD), scPPDM sets new state-of-the-art results across log fold-change recovery, delta correlations, explained variance, and DE-overlap. Representative gains include +36.11%/+34.21% on DEG logFC-Spearman/Pearson in UD over the second-best model. This control interface enables transparent what-if analyses and dose tuning, reducing experimental burden while preserving biological specificity.

Authors (6)

Zhaokang Liang

Shuyang Zhuang

Xiaoran Jiao

Weian Mao

Hao Chen

Chunhua Shen

Submitted

October 8, 2025

arXiv Category

q-bio.QM

arXiv PDF

Key Contributions

Introduces scPPDM, the first diffusion-based framework for single-cell drug-response prediction from scRNA-seq data. It couples pre-perturbation state and drug conditions in a unified latent space using non-concatenative GD-Attn. The model uses factorized classifier-free guidance for interpretable control over state preservation and drug-response strength, enabling tunable intensity via dose mapping. It achieves state-of-the-art results on challenging benchmarks.

Business Value

Accelerates drug discovery and development by enabling accurate prediction of cellular responses to various drugs and dosages, potentially reducing the need for extensive wet-lab experiments and enabling personalized medicine approaches.

Paper Metadata

Innovation Type

Algorithmic Development

Deployment Feasibility

Moderate. Requires specialized biological data (scRNA-seq) and computational resources for diffusion models. Integration into drug discovery pipelines is feasible.

Limitations Addressed

Addresses the challenge of predicting drug responses from single-cell data, especially under unseen conditions (covariates, drugs), and provides interpretable controls for the prediction process.

Performance Gains

+36.11%/+34.21% on DEG logFC-Spearman/Pearson in UD over the second-best model.

Technical Tags

diffusion modelssingle-cell RNA sequencingdrug response predictionperturbation predictionscPPDMGD-Attnclassifier-free guidanceunseen covariatesunseen drugsgene expression

Research Topics

Computational BiologySingle-Cell GenomicsDrug DiscoveryGenerative ModelsMachine Learning for Biology

Methods & Architectures

Diffusion modelsGD-Attn (non-concatenative)Factorized classifier-free guidanceConditional generation Diffusion ModelsGenerative Models

Applications & Tasks

Drug Discovery Genomics Precision Medicine Biotechnology Predicting cellular response to drugsModeling gene expressionGenerative modeling for biological data Predicting drug responseModeling single-cell perturbationsGenerating synthetic scRNA-seq data

Datasets & Benchmarks

Datasets

Tahoe-100M

Benchmarks

Tahoe-100M (under UC and UD regimes)

Log fold-change recoveryDelta correlationsExplained varianceDE-overlap

Related Fields

Computational BiologyBioinformaticsMachine LearningGenerative Models

Keywords

diffusion modelssingle-cell RNA sequencingdrug responseperturbation predictionscPPDMgenerative modelscomputational biologygene expressionprecision medicinebioinformaticsunseen covariatesunseen drugs

Academic Context

#Computational Biology#Single-Cell Genomics#Drug Discovery#Generative Models#Machine Learning for Biology

Commercial Potential

Potential Products

A platform for in-silico drug screeningA tool for predicting patient response to therapies

Target Industries

PharmaceuticalsBiotechnologyHealthcareResearch Institutions

Use Case Examples

Predicting the efficacy of new drug candidatesIdentifying optimal drug combinationsSimulating cellular responses to different treatments

Competitive Edge

First diffusion-based framework for this specific task, achieving state-of-the-art results and offering novel control mechanisms.

Market Opportunity

Large and growing market for AI in drug discovery and precision medicine.

Revenue Models

Software licensingService contractsPartnerships with pharma companies

Resource Requirements

Compute Needs

High (typical for diffusion models and large single-cell datasets)

Data Requirements

Single-cell RNA sequencing data, drug perturbation information.

Deployment Constraints

Requires expertise in single-cell data analysis and generative models; computational cost.

Scalability

Diffusion models can be computationally intensive, but the framework aims for practical application.

Regulatory Considerations

Data privacy (genomic data)Validation requirements for drug discovery tools

Production Readiness

Maturity Level

Research

Time to Market

2-4 years (for integration into drug discovery pipelines)

Patent Potential

Moderate (novel framework and application)

View Full Paper Back to Papers