Multimodal Learning Reveals Plants' Hidden Sensory Integration Logic

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: VOMO DONFACK, Kelly Larissa, González Doblas, Verónica, Morilla, Ian
Format: Recurso digital
Veröffentlicht: Zenodo 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866902257749983232
author VOMO DONFACK, Kelly Larissa
González Doblas, Verónica
Morilla, Ian
author_facet VOMO DONFACK, Kelly Larissa
González Doblas, Verónica
Morilla, Ian
contents <p>CoMM-BIP (Contrastive Multimodal Model with Biological Informed Priors) is an interpretable deep learning framework that integrates transcriptomic, metabolomic, phenomic, and environmental data to predict multi-sensory effector responses in plants. Key features:</p> <ul> <li> <p><strong>Biologically Guided Design</strong>: Hardcodes domain knowledge through:</p> <ul> <li> <p><em>BiologicalPriorLayer</em> (upweights known functional genes/metabolites)</p> </li> <li> <p><em>PathwayGuidedAttention</em> (models cross-modal interactions)</p> </li> </ul> </li> <li> <p><strong>Multimodal Fusion</strong>: Combines four data types using contrastive learning (NTXentLoss) and cross-modal attention.</p> </li> <li> <p><strong>State-of-the-Art Performance</strong>: Achieves 0.98 F1 score and 0.99 AUROC, outperforming unimodal baselines by 15-35%.</p> </li> <li> <p><strong>Interpretability</strong>: Provides attention matrices and SHAP values for biological insights.</p> </li> </ul> <p>The repository includes:</p> <ul> <li> <p>Pretrained PyTorch model (<code>*.pt</code> file)</p> </li> <li> <p>Training/evaluation scripts (<code>train.py</code>)</p> </li> <li> <p>Example notebooks for visualization</p> </li> <li> <p>Configuration files for reproducibility</p> </li> </ul> <p><strong>Applications</strong>:</p> <ul> <li> <p>Plant-pathogen interaction studies</p> </li> <li> <p>Multimodal biomarker discovery</p> </li> <li> <p>Interpretable ML for biological systems</p> </li> </ul> <p><strong>Technical Details</strong>:</p> <ul> <li> <p>Framework: PyTorch Lightning</p> </li> <li> <p>Language: Python 3.8+</p> </li> <li> <p>License: MIT</p> </li> </ul>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_16281076
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Multimodal Learning Reveals Plants' Hidden Sensory Integration Logic
VOMO DONFACK, Kelly Larissa
González Doblas, Verónica
Morilla, Ian
Machine learning
multimodal integration
plant-microbe interactions
interpretable AI
PyTorch
biological prior
plant systems biology
<p>CoMM-BIP (Contrastive Multimodal Model with Biological Informed Priors) is an interpretable deep learning framework that integrates transcriptomic, metabolomic, phenomic, and environmental data to predict multi-sensory effector responses in plants. Key features:</p> <ul> <li> <p><strong>Biologically Guided Design</strong>: Hardcodes domain knowledge through:</p> <ul> <li> <p><em>BiologicalPriorLayer</em> (upweights known functional genes/metabolites)</p> </li> <li> <p><em>PathwayGuidedAttention</em> (models cross-modal interactions)</p> </li> </ul> </li> <li> <p><strong>Multimodal Fusion</strong>: Combines four data types using contrastive learning (NTXentLoss) and cross-modal attention.</p> </li> <li> <p><strong>State-of-the-Art Performance</strong>: Achieves 0.98 F1 score and 0.99 AUROC, outperforming unimodal baselines by 15-35%.</p> </li> <li> <p><strong>Interpretability</strong>: Provides attention matrices and SHAP values for biological insights.</p> </li> </ul> <p>The repository includes:</p> <ul> <li> <p>Pretrained PyTorch model (<code>*.pt</code> file)</p> </li> <li> <p>Training/evaluation scripts (<code>train.py</code>)</p> </li> <li> <p>Example notebooks for visualization</p> </li> <li> <p>Configuration files for reproducibility</p> </li> </ul> <p><strong>Applications</strong>:</p> <ul> <li> <p>Plant-pathogen interaction studies</p> </li> <li> <p>Multimodal biomarker discovery</p> </li> <li> <p>Interpretable ML for biological systems</p> </li> </ul> <p><strong>Technical Details</strong>:</p> <ul> <li> <p>Framework: PyTorch Lightning</p> </li> <li> <p>Language: Python 3.8+</p> </li> <li> <p>License: MIT</p> </li> </ul>
title Multimodal Learning Reveals Plants' Hidden Sensory Integration Logic
topic Machine learning
multimodal integration
plant-microbe interactions
interpretable AI
PyTorch
biological prior
plant systems biology
url https://doi.org/10.5281/zenodo.16281076