Machine Learning To Predict Cell-Penetrating Peptides for Antisense Delivery.

Area of Science:

• Biotechnology
• Molecular Biology
• Drug Delivery

Background:

• Cell-penetrating peptides (CPPs) facilitate intracellular delivery of large molecules like proteins and oligonucleotides.
• Predicting effective CPP sequences for specific cargo remains a significant challenge.
• Phosphorodiamidate morpholino oligonucleotides (PMOs) are promising antisense therapeutics, notably for Duchenne muscular dystrophy.

Purpose of the Study:

• To develop a predictive model for identifying CPPs specifically for enhancing PMO delivery.
• To create and evaluate a library of PMO-CPP conjugates for training a computational model.
• To validate the predictive model's accuracy in identifying efficacious novel CPP sequences.

Main Methods:

• Synthesized 64 covalent PMO-CPP conjugates using known CPP sequences.
• Evaluated PMO activity using a fluorescence-based reporter assay.
• Developed a random decision forest classifier trained on experimental PMO activity data.
• Experimentally validated seven novel CPP sequences predicted by the model.

Main Results:

• Significant differences in CPP efficacy were observed between small-molecule fluorophore conjugation and PMO conjugation.
• The random decision forest model accurately predicted CPPs that enhance PMO activity.
• All seven computationally predicted positive CPP sequences demonstrated efficacy in experimental validation.
• One novel CPP sequence outperformed 80% of previously tested literature CPPs.

Conclusions:

• Machine learning algorithms can effectively identify peptide sequences with specific functional properties.
• Tailoring CPP sequences to the specific therapeutic cargo, like PMOs, is crucial for optimal delivery.
• This approach holds promise for advancing antisense oligonucleotide therapeutics and other large molecule delivery systems.