Artificial Neural Network Elucidates the Role of Transport Proteins in Rhodopseudomonas palustris CGA009 During Lignin Breakdown Product Catabolism

TL;DR

This study uses artificial neural networks to identify transport proteins in Rhodopseudomonas palustris that are important for breaking down lignin products.

Contribution

The study introduces a novel use of ANNs to identify key transport proteins involved in lignin catabolism in R. palustris.

Findings

ANNs achieved 94% accuracy with transcriptomics data and 96% with proteomics data.

Eight transport proteins consistently influenced growth across lignin breakdown product conditions.

Using only these eight proteins, ANNs maintained 86% proteomics and 76% transcriptomics accuracy.

Abstract

Background: Rhodopseudomonas palustris is a metabolically versatile bacterium with significant biotechnological potential, including the ability to catabolize lignin and its heterogeneous breakdown products. Understanding the molecular determinants of growth on lignin-derived compounds is essential for advancing lignin valorization strategies under both aerobic and anaerobic conditions. Methods: R. palustris was cultivated on multiple lignin breakdown products (LBPs), including p-coumaryl alcohol, coniferyl alcohol, sinapyl alcohol, p-coumarate, sodium ferulate, and kraft lignin. Condition-specific transcriptomics and proteomics datasets were generated and used as input features to train machine-learning models, with experimentally measured growth rates as the prediction target. Artificial Neural Networks (ANNs), Random Forest (RF), and Support Vector Machine (SVM) models were evaluated and compared. Permutation feature importance analysis was applied to identify genes and proteins most influential for growth. Results: Among the tested models, ANNs achieved the highest predictive performance, with accuracies of 94% for transcriptomics-based models and 96% for proteomics-based models. Feature importance analysis identified the top twenty growth-associated genes and proteins for each omics layer. Integrating transcriptomic and proteomic results revealed eight key transport proteins that consistently influenced growth across LBP conditions. Re-training ANN models using only these eight transport proteins maintained high predictive accuracy, achieving 86% for proteomics and 76% for transcriptomics. Conclusions: This study demonstrates the effectiveness of ANN-based models for predicting growth-associated genes and proteins in R. palustris. The identification of a small set of key transport proteins provides mechanistic insight into lignin catabolism and highlights promising targets for metabolic engineering aimed at improving lignin utilization.

Figures