Prediction of carbon and nitrogen source preferences in microbial metabolism using protein sequence data

Microbial precision cultivation technology holds significant application value in the field of environmental pollutant remediation. Precise quantification of microbial carbon and nitrogen requirements is critical for optimizing culture conditions and enhancing microbial growth and productivity. This study aims to explore the intrinsic relationship between microbial protein sequences and their specific nutritional requirements (e.g., the types of carbon and nitrogen sources as well as the optimal carbon-to‑nitrogen (C/N) ratio) using deep learning algorithms. A total of 432 microbial species and 61 culture media formulations were collected from authoritative databases, including Ensembl Bacteria, DSMZ, and NCBI Protein. For data analysis, microbial protein sequences were converted into high-dimensional numerical feature matrices using the Position-Specific Scoring Matrix (PSSM) and Pseudo Position-Specific Scoring Matrix (PsePSSM), followed by dimensionality reduction. Multiple machine learning algorithms were employed to construct predictive models for microbial C/N source utilization. Among the classification tasks, C/N ratio prediction performed best, with an accuracy of 99.60 %, followed by carbon source prediction with an accuracy of 82.76 % and nitrogen source prediction with an accuracy of 70.05 %, suggesting a strong correlation between microbial C/N requirements and their associated protein sequences. Furthermore, model interpretability was enhanced using the SHapley Additive exPlanations (SHAP) framework to analyze feature contributions. The primary contribution of this study lies in proposing an integrated framework that combines protein function annotation with sequence-based feature extraction for predicting microbial nutritional requirements, thereby offering new insights for optimizing microbial culture conditions.