A method for predicting polypeptide activity and toxicity

By using a hybrid Transformer-CNN feature encoding network to collaboratively model peptide activity and toxicity within a unified model framework, the fragmentation problem in existing technologies is solved, enabling joint prediction of peptide activity and toxicity. This improves the accuracy and stability of the predictions and is suitable for high-throughput screening and safety assessment of peptide candidate molecules.

CN122201443APending Publication Date: 2026-06-12CHINA PHARM UNIV
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHINA PHARM UNIV
Filing Date
2026-03-25
Publication Date
2026-06-12

AI Technical Summary

⚠Technical Problem

Existing technologies suffer from fragmented targets for predicting peptide activity and toxicity, limited granularity of sequence feature characterization, and insufficient model stability, making it difficult to collaboratively model within a unified framework. Consequently, the comprehensive guiding significance of the prediction results in peptide screening is limited.

⚗Method used

We employ a multi-scale sequence feature representation and deep feature fusion mechanism, and use a hybrid Transformer-CNN feature encoding network to co-model peptide activity and toxicity within a unified model framework. By combining residue-level semantic representation, physicochemical property features and n-gram sequence statistical features, we construct the local structural patterns and global dependencies of peptide sequences.

🎯Benefits of technology

It improves the accuracy and stability of peptide activity and toxicity prediction, enhances the generalization ability of the model, provides a unified and reliable evaluation basis for peptide screening and optimization, and improves the screening efficiency of peptide candidate molecules.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

The application discloses a polypeptide activity and toxicity combined method, comprising the following steps: obtaining a polypeptide amino acid sequence sample and corresponding activity annotation information and toxicity annotation information, preprocessing the polypeptide amino acid sequence, and constructing a polypeptide amino acid sequence dataset; constructing a multi-scale feature representation, and uniformly modeling residue-level semantic features, physicochemical property features and sequence statistical features, inputting a mixed feature coding network to extract global semantic information and local structure patterns of the polypeptide; constructing a multi-task prediction structure in a shared feature space, synchronously outputting activity scores and toxicity classifications of the polypeptide, and realizing joint prediction and screening of polypeptide activity and toxicity; the application can simultaneously obtain activity scores and toxicity probability prediction results of the polypeptide in the same prediction process, thereby more intuitively evaluating the functional effect and safety of the polypeptide, and significantly improving the accuracy and prediction efficiency of polypeptide screening.
Need to check novelty before this filing date? Find Prior Art