Introduction
In 2024 the Nobel Prize in Physics was awarded to Geoffrey Hinton and John J. Hopfield to recognize their foundational discoveries that enabled ML based on artificial neural networks. Those fundamental developments happened in the 1980s when computational power and access to data for training models was still limited. Advances in computation and deep learning algorithms for artificial neural networks driven by applications in physics changed dramatically and led to fundamental breakthroughs in AI/ML in the 2000s. Today, machines can execute tasks which partly outperform human experts and even world champions in their fields like in chess and the ancient Chinese strategy board game GO.
The life and materials science community also started to embrace the benefits of AI/ML around the same period to generate faster, better and/or more insights from rapidly accumulating data. Many of those insights escaped our human intuition and understanding before due to the complexity of underlying mechanisms of the studied biological systems and materials.
The breakthroughs of AI/ML that have been demonstrated in the Life Sciences and Materials Science field are very impressive. Many of these tools speed up discoveries and already deliver tangible innovations which are covered below.
AI for Life Science Applications
Biotechnology Applications
Biotechnology deals with highly complex biological systems which delivers amazing solutions to various industries. Synthetic Biology promises transformative solutions based on systems biology tools which provide more and better data to learn from. This field also benefits from increasingly efficient engineering tools such as Directed Evolution and Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)
technology which both were awarded separate Nobel Prizes in Chemistry in 2018 and 2020 respectively (1, 2).
The introduction of AI/ML tools will advance biotechnology and systems biology by providing new insights from rapidly accumulating genomics, proteomics, metabolomics and protein structure data. One of the first breakthrough was the high quality prediction of protein structures from amino acid sequences and computational protein design strategies which earned Demis Hassabis, John Jumper and David Baker the 2024 Nobel Prize in Chemistry (3).
The scientific field rapidly picked up Alphafold 2 (4) as initial standard for structure predictions which is tremendously expanding our insights from protein structures. For example, Hassabis assessed the entire human proteome (98,5% of all human proteins) and made all structural models available to the scientific community. These initial AI based models already have a profound impact on the structure based (rational) enzyme engineering space which was often limited by the access to reliable structural models. Additional progress has been made with improved AI models such as Alphafold3 (5) and the open source Boltz-1 model (6) which can predict the interaction of different biomolecules (protein, DNA, RNA, peptides) and provide structures of such protein complexes with unprecedented accuracy.
In addition to structure prediction models, AI/ML models are emerging which try to learn functional features from protein sequences such as for example enzyme classes (7), stability (8), expression (9), substrate scope (10) affinity and catalytic efficiency features (11).
Some predictions like enzyme class categorization and stability are already quite reliable whereas the predictive power of other models are still insufficient to support rational enzyme engineering projects. The lack of predictive power and generalization for some features is not surprising as proteins are flexible and impacted by many biomolecular interactions which are highly context dependent (e.g., pH, temperature, matrix) and are not fully embedded in sequence information.
The latest structure prediction tools now include molecular dynamics and can predict conformational changes of proteins (12). Furthermore AI/ML tools recently introduced simulate full-atom enzymes dynamics with claimed ab initio accuracy requiring orders of magnitude lower computational power than traditional methods (13). As there are still some remaining challenges in rational protein design (14) alternative strategies such as smart library designs (15) and directed evolution (16) also benefit from AI models which can efficiently guide enzyme engineering cycles. In addition to the above described impact on enzyme engineering, de novo design of enzymes for novel chemistries (17), as well as for in silico enzyme discovery (18) are rapidly advancing our tools. All these developments will further unlock the great potential of enzymes, synthetic biology and industrial biotechnology as a whole.
Pharma Applications
The previous section was dealing with AI/ML tools which are starting to be used for predicting functional features of highly complex biomolecules such as enzymes. Understanding the function and dysfunction of human bodies is several orders of magnitude more complex and a big challenge for the pharmaceutical industry. This challenge is visible by a low success rate and huge costs to bring a new drug to the market. Within drug discovery, many tools are developed that predict for example binding properties of inhibitor’s binding energy, assess how easy or difficult a molecule can be synthesized, their pharmacokinetic profile property predictions such as solubility, bioactivity, or ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties for given molecular structures. For a recent review on AI-assisted drug screening approaches see for example Singh et al. (19).
These in-silico screening approaches have also been combined in fully automated generative AI pipelines which delivered more that 25-AI assisted drug candidates under clinical investigation (20).
However there are also setbacks reported in clinical trials with a few of these candidates. Whether AI/ML will increase the success rate and reduces the cost of drug development is still too early to be judged. The speed at which new tools and applications are developed and the rapidly expanding and publicly accessible codes, models as well as data is promising a bright future of AI/ML which will revolutionize the Pharmaceutical industry and, most importantly, will finally lead to improved patient outcomes.
Materials Science Applications
Just like in Life Sciences, there is significant hope that AI/ML will have a transformative impact on Materials Science innovations. The promise is predicting materials properties from chemical compositions and processing parameters to accelerate the discovery and development of new materials at reduced experimental burden.
Early AI/ML algorithms have been successfully developed to predict a wide range of mechanical, thermal, and electrical properties, such as Young’s modulus (21), melting temperature (22), and conductivity (23). These AI-driven predictions are crucial for designing advanced materials tailored for specific applications.
For instance, AI has been instrumental in the design of materials for next-generation batteries (24), where accurate predictions of ionic conductivity, stability, and capacity are essential. Similarly, AI models have been applied to develop metallic alloys with optimized strength, ductility, and corrosion resistance (25). In the realm of catalysis, ML algorithms have facilitated the identification of catalysts (26) with enhanced activity, selectivity, and stability, which are vital for chemical synthesis processes.
Very recent advancements in AI for materials science have been marked by groundbreaking contributions from big AI players that already transformed other fields. One notable example is DeepMind’s use of graph neural networks trained on high-throughput quantum-mechanics-based first-principles calculations (27). Their approach, known as Graph Networks for Materials Exploration (GNoME), led to the discovery of over 2.2 million stable crystalline materials. These materials, with their diverse structural and compositional properties, open new avenues for developing innovative technologies in energy storage, electronics, and photonics.
In addition to predictive models, generative models are making significant strides in materials discovery. Microsoft has developed MatterGen, a generative model that can design novel materials with specific desired properties (28). MatterGen excels in creating materials with tailored chemistry, symmetry, and mechanical, electronic, and magnetic properties. This capability enables the rapid prototyping of materials for applications ranging from superconductors to lightweight composites and multifunctional materials.
AI’s role in Materials Science as well as in the Life Sciences extends beyond mere prediction and in-silico generation. It can also facilitate the generation of data, optimization of experimental processes and the interpretation of complex data. By integrating AI with automated high-throughput experimentation, advanced characterization techniques, and optimization techniques such as Bayesian Optimization, researchers started to develop so called Self Driving Labs (29). While this sounds futuristic, first commercial solutions are developed (30).
Conclusion
The integration of AI/ML into Life Sciences and Materials Science is poised to drive significant advancements, enabling the rapid discovery and deployment of new products with unprecedented properties and functionalities. The expectations surrounding AI/ML are substantial, yet sometimes overly ambitious. Practitioners may experience disappointments when the predictive power of AI models fails to meet experimental results. This gap is often attributed to the insufficient quantity and quality of data used for training these models.
In scenarios where data is scarce, and ab initio calculations remain computationally too expensive, scientists must turn to alternative tools for efficient data generation and optimization tasks. While the future role of Self Driving Labs in addressing these challenges is still uncertain, advanced optimization algorithms, such as Bayesian Optimization, are already proving invaluable. These tools help scientists streamline their efforts, reducing the experimental burden while improving the efficacy of their solutions.
As AI/ML technologies continue to evolve, the growing expectations will be progressively realized, providing scientists with powerful tools and methodologies to deliver faster and more effective solutions for the Life Science and Material Science industry, ultimately benefiting both human as well as our planet’s health.
References and Notes
1. The Nobel Prize in Chemistry 2018 [Internet]. Available from: https://www.nobelprize.org/prizes/chemistry/2018
2. The Nobel Prize in Chemistry 2020 [Internet]. Available from: https://www.nobelprize.org/prizes/chemistry/2020
3. The Nobel Prize in Chemistry 2024 [Internet]. Available from: https://www.nobelprize.org/prizes/chemistry/2024
4. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with Alphafold. Nature. 2021;596(7873):583–589.
5. Abramson J, Adler J, Dunger J, Evans R, Green T, Pritzel A, et al. Accurate structure prediction of biomolecular interactions with Alphafold 3. Nature. 2024;pages 1–3.
6. Wohlwend J, Corso G, Passaro S, Reveiz M, Leidal K, Swiderski W, et al. Boltz-1: Democratizing biomolecular interaction modeling. bioRxiv. 2024.11.19.624167. doi: https://doi.org/10.1101/2024.11.19.624167
7. Yu T, Cui H, Li JC, Luo Y, Jiang G, Zhao H. Enzyme function prediction using contrastive learning. Science. 2023;379(6639):1358–1363.
8. Ming Y, Wang W, Yin R, Zeng M, Tang L, Tang S, et al. A review of enzyme design in catalytic stability by artificial intelligence. Brief Bioinform. 2023;24(3):bbad065. doi: https://doi.org/10.1093/bib/bbad065
9. Gilliot P-A, Gorochowski TE. Transfer learning for cross-context prediction of protein expression from 5’UTR sequence. Nucleic Acids Res. 2024;52:e58. doi: https://doi.org/10.1093/nar/gkae491.
10. Lercher MJ, et al. A general model to predict small molecule substrates of enzymes based on machine and deep learning. Nat Commun. 2023;14:2787.
11. Alazmi M. Enzyme catalytic efficiency prediction: employing convolutional neural networks and XGBoost. Front Artif Intell. 2024;7:1446063.
12. Song X, Bao L, Feng C, Huang Q, Zhang F, Gao X, et al. Accurate prediction of protein structural flexibility by deep learning integrating intricate atomic structures and Cryo-EM density information. Nat Commun. 2024;15:5538.
13. Wang T, He X, Li M, Li Y, Bi R, Wang Y, et al. Ab initio characterization of protein molecular dynamics with AI2BMD. Nature. 2024;635:1019.
14. Reardon S. Five tasks that still challenge protein designers. Nature. 2024;635:246–248.
15. Patsch D, Eichenberger M, Voss M, Bornscheuer UT, Buller RM. LibGENiE: A bioinformatic pipeline for the design of information-enriched enzyme libraries. Comput Struct Biotechnol J. 2023;21:4488–4496.
16. Li F-Z, Yang J, Johnston KE, Gürsoy E, Yue Y, Arnold FH. Evaluation of machine learning-assisted directed evolution across diverse combinatorial landscapes. bioRxiv. 2024.10.24.619774. doi: https://doi.org/10.1101/2024.10.24.619774.
17. Winnifrith A, Outeiral C, Hie BL. Generative artificial intelligence for de novo protein design. Curr Opin Struct Biol. 2024;86:102794.
18. Ariaeenejad S, Gharechahi J, Foroozandeh Shahraki M, Atanaki FF, Han J-L, Ding X-Z, et al. Precision enzyme discovery through targeted mining of metagenomic data. Nat Prod Bioprospect. 2024;14:7.
19. Singh S, Gupta H, Sharma P, Sahi S. Advances in artificial intelligence (AI)-assisted approaches in drug screening. Artif Intell Chem. 2024;2:100039.
20. Gangwal A, Lavecchia A. Unleashing the power of generative AI in drug discovery. Drug Discov Today. 2024;29(6):1-24.
21. Yang K, Xu X, Yang B, Cook B, Ramos H, Krishnan N, et al. Predicting the Young’s modulus of silicate glasses using high-throughput molecular dynamics simulations and machine learning. Sci Rep. 2019;9(1):1-11.
22. Qu N, Liu Y, Liao M, Lai Z, Zhou F, Cui P, et al. Ultra-high temperature ceramics melting temperature prediction via machine learning. Ceram Int. 2019;45(15):18551-18555.
23. Chen L, Tran H, Batra R, Kim C, Ramprasad R. Machine learning models for the lattice thermal conductivity prediction of inorganic materials. Comput Mater Sci. 2019;170:109155.
24. Lv C, Zhou X, Zhong L, Yan C, Srinivasan M, Seh ZW, et al. Machine learning: an advanced platform for materials development and state prediction in lithium-ion batteries. Adv Mater. 2022;34(25):2101474.
25. Xu Z-W, Liu X-M, Zhang K. Mechanical properties prediction for hot rolled alloy steel using convolutional neural network. IEEE Access. 2019;7:47068-47078.
26. McCullough K, Williams T, Mingle K, Jamshidi P, Lauterbach J. High-throughput experimentation meets artificial intelligence: a new pathway to catalyst discovery. Phys Chem Chem Phys. 2020;22(20):11174-11196.
27. Merchant A, Batzner S, Schoenholz SS, Aykol M, Cheon G, Cubuk ED. Scaling deep learning for materials discovery. Nature. 2023;624:80-85.
28. Zeni C, et al. MatterGen: a generative model for inorganic materials design. arXiv preprint. 2023. arXiv:2312.03687.
29. Bayley O, Savino, E, Slattery, A, Noel, T. Autonomous chemistry: Navigating self-driving labs in chemical and material sciences. Matter. 2024; 7(7): 2382-2398
30. https://telescopeinnovations.com/self-driving-labs/
