There is enough room for AI in proteomics to make an impact on drug discovery and disease prevention
AI in proteomics is gaining new significance in the event of a pandemic-ridden world. The research community is burning out to find a new vaccine every time the virus mutates. Proteins are an important part of functional molecules in the cell. They are affected by the mutations that occur in the DNA. The viruses mutate because proteome or the set of proteins tend to fluctuate with time. Therefore, AI in proteomics research constitutes an important segment of Drug research. It is crucial to understand the protein structure, its functions, and its interactions.
DeepMind and EMBL recently released 350,000 protein structures computed by the AlphaFold system. These were computed using DeepMind’s algorithms to find more protein structures. Understanding these structures is key to understanding how they function.
How much AI can predict?
For years scientists were using computer models for grasping the mechanism of how protein strands fold and end up in their respective 3D shapes. But these models when applied to other proteins would fail to show similar results. Deep learning captures the interactions between different proteins learned from the data pertaining to previous interactions. However, accurate results are possible only when the right inputs in terms of related and unrelated sequences are fed to the algorithms to generate accurate results using AI in proteomics. And also, they could only predict the possible structures but not the reason for a particular structure. As of now, these AI-enabled predictions stand as defining techniques in the field of science and research.
AI in model development
Protein engineering is an evolving area riding on the back of AI-enabled model development capabilities. It aims to generate new proteins with improved functions. Directed evolution is one crucial strategy adopted by scientists in model development. However, as it involves a number of iterative cycles and high throughput screen selection, it has its own limitations. ML algorithms can bridge this gap by learning the sequence-function relationship from sequence and screening data. It can bypass the requirement of local optimum conditions needed for a set of sequences with the help of available sequence and screening data. ECNet, an evolutionary context-integrated deep learning framework can predict higher-order mutations using lower order mutant data, further pushing AI in proteomics into higher realms.
ML in biomarker identification
Mass spectroscopy aided by ML techniques can prove to be crucial in biomarker identification. which are used to identify a biological situation or predict a biological activity. Huge data sets generated in mass spectroscopy with computational intervention reduce drastically the processing time. Identification of unique biomarkers using an algorithmic approach aids accurate drug discovery for rare diseases. In a recent study at Martin Luther University and the European Molecular Biology Laboratory, researchers have developed an AI-based method for analyzing cryo-electron microscopy data to study the complex of proteins simultaneously and directly in the cells. This is critical development because better results can be achieved if clusters of proteins are studied in their real environment.