Methods, electronic devices, computer-readable storage media, and systems for predicting microbial phenotypes based on genomes
By combining DNABERT-2 with K-mer statistical methods and CCA-PLS analysis, and integrating the deep semantic and local structural features of the genome sequence, this method solves the problems of high cost, low efficiency, and insufficient accuracy in the prediction of probiotic acid and bile salt tolerance in existing technologies, and achieves efficient and accurate strain screening and prediction.
Patent Information
- Application Number
- CN202610571371.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for screening and predicting the acid and bile salt tolerance of probiotics suffer from high cost, low efficiency, poor stability, and insufficient prediction accuracy and generalization ability. In particular, traditional methods cannot effectively integrate global semantics and local statistical features, leading to feature redundancy and noise amplification.
By combining a DNABERT-2 pre-trained Transformer model with the K-mer statistical method, and fusing deep semantic features and local structural features of the genome sequence through the CCA-PLS analysis method, a global-local multi-scale feature system was formed to predict the acid and bile salt tolerance of microorganisms.
It achieves high-throughput, low-cost strain screening, improves prediction accuracy and generalization ability, can complete screening without in vitro experiments, significantly enhances the model's discriminative ability and stability across datasets, and is suitable for parallel screening of multiple phenotypes.
Smart Images
Figure CN122417147A_ABST