SDPPP (Silica Deposition Protein Prediction Pipeline): A transfer learning–based pipeline for predicting candidate silica deposition proteins independent of sequence homology

SDPPP (Silica Deposition Protein Prediction Pipeline) is a command-line tool for identifying putative silica deposition–related proteins from amino-acid sequences. It leverages sequence embeddings generated by the ESM-2 protein language model (esm2_t33_650M_UR50D) and applies a Random Forest classifier to perform supervised learning and inference. SDPPP supports two main workflows: (i) training a model from user-provided positive and negative FASTA datasets, and (ii) predicting silica deposition proteins from a candidate FASTA file using either a newly trained model or the provided pretrained classifier. The pipeline outputs tabular prediction results, including probability scores, binary labels (based on a configurable threshold), and the corresponding embedding features to facilitate downstream analysis and reproducible screening.