NpSearch: Search for Neuropeptides

This repository contains the datasets, scripts, and analyses for the Laboratory of Bioinformatics II course project, focusing on the prediction of secretory signal peptides.

A comprehensive protein analysis platform that combines motif detection with 3D structure validation using AlphaFold2 confidence scores. Designed for cancer biology research, drug discovery, and functional genomics.

To predict cryptic cleavage sites in proteins with non-canonical signal peptides

A re-built convolutional neural network from paper「DeepSig: deep learning improves signal peptide detection in proteins」

A machine learning pipeline for eukaryotic signal peptide prediction — from UniProtKB data collection to a CNN-LSTM classifier on ESM-2 embeddings. Implements and benchmarks three methods: a von Heijne PSWM baseline, an SVM with biochemical features, and a deep learning model (F1=0.962, ROC-AUC=0.998).

Origin mechanisms of SAARs – use case of L-SAARs in signal peptides among higher Eukaryotes

Signal peptide prediction using deep learning. SandwichSP uses a CNN-LSTM-CRF architecture with ProtT5 embeddings to predict signal peptides and their cleavage sites in protein sequences.