Predicting Lactobacillus delbrueckii subsp. bulgaricus-Streptococcus thermophilus interactions based on a highly...

Area of Science:

• Microbiology
• Bioinformatics
• Dairy Science

Background:

• Lactobacillus delbrueckii subsp. bulgaricus (L. bulgaricus) and Streptococcus thermophilus (S. thermophilus) are crucial starter cultures in milk fermentation, significantly influencing dairy product quality.
• Traditional methods for screening starter culture interactions are inefficient, necessitating advanced computational approaches.
• Existing artificial intelligence models often require extensive labeled data, posing a challenge for predicting interactions with limited samples.

Purpose of the Study:

• To develop and validate a semi-supervised learning framework for predicting Lactobacillus-Streptococcus interactions (LbStI) using genomic data.
• To address the limitations of supervised learning in bioinformatics by utilizing a small number of labeled samples for interaction prediction.
• To identify key genomic features and pathways associated with the mutualistic relationships between L. bulgaricus and S. thermophilus.

Main Methods:

• A semi-supervised learning framework was developed, integrating a co-clustering model (KEGG dataset) and a Laplacian regularized least squares model (K-mer and gene composition analysis).
• Genomic data from 362 isolates (181 of each species) were utilized for model training and prediction.
• The integrated model's predictions were validated through milk fermentation experiments with randomly selected isolate combinations.

Main Results:

• The developed semi-supervised learning framework achieved a high precision rate of 85% in predicting LbStI, as confirmed by milk fermentation experiments.
• The study identified specific biosynthetic pathways (cysteine, riboflavin, teichoic acid, exopolysaccharides) and ATP-binding cassette transport systems as potentially contributing to the mutualistic relationship.
• The model provides a valuable resource for screening dairy starter cultures and understanding bacterial interactions.

Conclusions:

• The semi-supervised learning framework offers an efficient and accurate method for predicting dairy starter culture interactions, overcoming limitations of traditional and supervised methods.
• Genomic insights suggest specific metabolic pathways and transport systems are key to the symbiotic relationship between L. bulgaricus and S. thermophilus.
• This research provides a valuable computational tool and foundational data for optimizing dairy fermentation processes and starter culture selection.