Fighting fungal infections is a major challenge in global healthcare, especially with the rise of drug-resistant fungi and diminishing effectiveness of current treatments. While traditional approaches have made some progress in discovering antifungal compounds, the lack of efficient screening tools for short peptides has significantly slowed the research process.

Our team is pioneering the use of machine learning and the DeBERTa language model to predict and screen peptide sequences with potential antifungal activity. This innovative approach accelerates the discovery of antifungal peptides and marks a new milestone in AI-powered biosynthesis research.

With increasing antibiotic resistance, fungal infections are a growing public health threat worldwide. The World Health Organization (WHO) has classified certain fungi as high-priority pathogens due to drug resistance. Current antifungal treatments are limited, and resistance to these drugs has led to fewer viable treatment options.

Existing drugs are costly and take years to develop, making it difficult to respond quickly to drug-resistant infections. Our team aims to tackle this issue with a machine-learning solution that accelerates the screening and prediction of antifungal peptides, with a focus on discovering new, effective short peptides for Candida albicans treatment.

Candida albicans is a common fungus found in the human body, typically residing in the mouth, intestines, skin, and reproductive system. Normally, it is part of the body’s symbiotic microbiota and does not cause harm. However, when the immune system is weakened, or the body’s microbial balance is disrupted, Candida albicans can multiply rapidly, leading to infections known as candidiasis.

• Mouth and throat infections
• Genital infections
• Skin and nail infections
• Systemic infections

We employed the pre-trained DeBERTa language model and machine learning techniques to screen peptides with potential anti-Candida albicans activity from the NCBI database. By leveraging millions of biological sequences, we identified short peptides with significant antifungal activity, suitable for drug development.

1. Qualitative and Quantitative Two-Step Model: This innovative two-step model allows us to qualitatively identify antifungal activity regions in peptide sequences and further evaluate the minimum inhibitory concentration (MIC) against target fungi through quantitative analysis.

2. Efficient Sequence Screening: Using machine learning, we rapidly screened biological sequences for antifungal short peptides. Combined with quantitative analysis tools, we scored the selected peptides based on their biological stability and other key factors.

3. Application of Language Models: The DeBERTa model’s pre-training capabilities enable us to effectively screen and classify large sets of biological sequences. This is the first time that natural language processing has been used to predict antifungal peptides, paving the way for synthetic biology powered by machine learning.

Looking ahead, we plan to refine our models, expand the screening range of antifungal peptides, and further investigate their mechanisms of action. We also aim to collaborate with biopharmaceutical companies to bring peptide-based drugs to market, addressing the growing issue of drug-resistant fungal infections.

By combining machine learning and language models for antifungal peptide screening, we have provided a new solution for combating fungal infections. This breakthrough not only sets a precedent in biosynthesis research but also lays the groundwork for future AI-driven drug development.

Reference: World Health Organization. (2023). Fungal priority pathogens list to guide research, development and public health action. WHO