In recent years, the application of artificial intelligence (AI) in the field of biology has seen significant advancements, culminating in the recognition of AI's capabilities with the award of a Nobel Prize for Chemistry in 2024. This accolade was bestowed upon University of Washington biochemist David Baker and researchers Demis Hassabis and John Jumper from DeepMind for their groundbreaking work on solving the protein-folding problem, a long-standing challenge in biological sciences.
The protein-folding problem, which involves determining the three-dimensional structure of proteins from their linear amino acid sequences, had perplexed scientists for nearly two decades. The journey to its resolution began with the establishment of the Critical Assessment of Structural Prediction (CASP) competition in 1994, aimed at fostering collaborative efforts in structural biology. Baker's team developed the Rosetta software for protein energy modelling in 1998, and later launched the game Foldit to engage the public in protein structure determination. DeepMind’s introduction of the AlphaFold program in 2018 marked a significant breakthrough by accurately predicting protein structures, exploiting a dataset of over 100,000 known protein sequences and their structures. By 2020, AlphaFold2 had been heralded as a game-changer, with experts stating that the protein-folding problem was largely solved. This achievement was formalised in 2024 with recognition from the scientific community and the awarding of the Nobel Prize.
The success of AI in biology is not confined solely to protein structure prediction. It is revolutionising diverse areas such as drug design and cellular biology. For example, AI models are being increasingly utilised to create spatiotemporal maps of cells, analyse cellular images for morphological changes indicative of disease, and assess the effectiveness of new pharmaceuticals in preventing disease progression, thus streamlining the drug discovery process.
David Baker's ongoing efforts involve using deep learning models to engineer de novo proteins—proteins designed from scratch that are better suited for solving contemporary challenges compared to their natural counterparts. Recently, Baker's team successfully created luciferase enzymes capable of binding to synthetic luciferin to emit light, which holds promise for advanced imaging techniques in biological research.
Significantly, the rise of antimicrobial resistance has prompted researchers to harness AI in the development of novel antibiotics. Jon Stokes and his team at McMaster University have developed SyntheMol, a generative AI model that aids in creating new small molecules with antibacterial properties, particularly targeting the ESKAPE pathogen Acinetobacter baumannii, which poses a substantial global health risk. Although the newly designed compounds have yet to be tested in human trials, initial in vitro results indicate their potential effectiveness against other resistant microbes.
Complementing these advancements, new AI models inspired by neural networks are proving adept at processing complex biological data. These artificial neural networks (ANNs) can reveal intricate patterns within datasets that may elude human analysis, thus enabling researchers to automate less complex tasks and focus on higher-level scientific inquiries.
Moreover, breakthroughs in language models, akin to those seen in AI applications like ChatGPT, have extended into biology. For instance, researchers at the University of Texas at Austin, led by Alexander Huth, have designed a model capable of inferring thoughts from MRI data. While this technology is not yet applicable across varied subjects, it has generated insights into brain function, hinting at potential communication tools for individuals with speech impairments.
Entwined in these developments is the evolution of scGPT—a generative pre-trained transformer introduced by Bo Wang and his team at the University of Toronto. This AI model enhances the analysis of single-cell RNA sequencing data by predicting gene expression profiles more accurately than traditional methods. Originally focused on immune and bone marrow cells, the model has been adapted to investigate a broader range of cell types, presenting future possibilities for profound biological discoveries.
Despite the expansive potential of AI in understanding biological systems and creating novel therapies, experts stress the importance of leveraging such technology with caution. The success of AI applications in biology relies heavily on the researchers' comprehensive knowledge in the field, as AI remains a tool that complements, rather than replaces, human expertise. Consequently, the field continues to explore and refine AI models for a multitude of applications, heralding a new era in biological research and therapeutic development.
Source: Noah Wire Services