An idiot's guide to lead optimisation for proteins (magnusross.github.io)

A new blog post introduces the Cradle-1 pipeline for lead optimization in protein drug design, emphasizing its significance for the AI/ML community. Lead optimization is a critical stage in drug development, where researchers refine a partially effective molecule to enhance its functionality. The author shares insights from experienced peers, explaining that proteins, composed of amino acids, must fold correctly to perform vital biological functions. With the help of machine learning, particularly using a transformer-based model, researchers can predict the viability of amino acid substitutions in proteins, streamlining the optimization process. Cradle, a biotech startup, has developed a machine learning platform that integrates both computational insights and experimental validation by maintaining its own wet lab. Their approach leverages a massive dataset of natural proteins, using techniques like masked language modeling and evolutionary fine-tuning (evotuning) to focus on specific functional improvements by exploring evolutionarily related proteins. By employing these advanced methods, Cradle aims to enhance the efficiency of lead optimization, representing a significant leap forward in the intersection of AI and biopharmaceutical development, which could lead to more effective therapeutics.