Generative AI for Biomolecular Design: Small Molecules, Proteins, and RNA

Advancements in Biomolecular Design

• 💡 Generative AI is transforming drug discovery and protein engineering, building on successes like AlphaFold for structure prediction.
• 🎯 The core challenge is to design biomolecules with desired properties, using generative models, scoring models, and experimental validation.
• 🔑 Research spans small molecules, proteins, and RNA, each with unique design considerations and applications.

De Novo Antibiotic Discovery

• ⚠️ Antibiotic resistance is a critical global health threat, necessitating the discovery of novel antibiotic classes.
• 🧪 A generative AI workflow, combining GNN-based property prediction (ChemProp) and fragment-based generation (JT-VAE), was developed.
• ✅ This approach successfully designed novel antibiotics for Gonorrhea (NG1) and MRSA (DN1), demonstrating improved selectivity and in vivo efficacy.

Enhancing Protein Binder Design

• 🧬 Designing protein binders is crucial for therapeutics, but existing methods like BindCraft face challenges with non-smooth optimization landscapes.
• 📊 Introduced PTM energy, a probabilistic score derived from AlphaFold's confidence heads, offering a smoother and more predictive metric for binding.
• 🚀 BindEnergyCraft, utilizing PTM energy, achieved 5-10% improvement in binder design and reduced structural clashes compared to standard methods.

Innovative RNA Design with RNA Flow

• 🧩 RNA design offers versatile functions (aptamers, riboswitches) but traditional methods are costly and de novo design is challenging.
• ⚡ Developed RNA Flow, a flow matching model that generates RNA structures and sequences without expensive fine-tuning of large folding models like RoseTTAFold-RNA.
• 🌱 RNA Flow demonstrated improved sequence recovery and structural accuracy, successfully designing aptamers for GRK2 with critical binding motifs.