In this section we will cover the importance of programming and AI in chemistry and biochemistry, highlighting some notable examples in the literature. We will also summarize the course content.

Programming or coding involves the writing of a sequence of instructions, these instructions can then be executed by a computer. Programming allows us to automate tasks, making things that would take a human a very long time to do possible.

The true definitions of Artificial Intelligence (AI) and Machine Learning (ML) differ slightly from how they are often used colloquially. We will first review the accepted definitions:
AI - a program that appears to mimic human thinking, it can respond to input and provide an output, it can include both programs that learn from data and may act in ways that humans don't understand as well as hard coded rules or a mix of both. For example if you were developing a self-driving car AI, you may want it be trained on data of how drivers behave so that it acts in a similar way, but for safety you may want to add additional hard coded rules - for example to brake if there is an object very close to the front of the car. This is a broad category that includes any sort of program that appears to mimic human thinking.
ML - machine learning is a subset of AI, it includes programs that learns outputs/decisions based on data. It excludes hard coded rules like the ones discussed above. As we will see, there are many types of ML models some that are relatively simple and others, called artificial neural networks (ANNs), that are inspired by the brain.
Colloquially, people often use AI and ML interchangeably or limit use of AI to more advanced programs, such as large language models like ChatGPT while sometimes only using ML to describe simplistic models and not using it to describe ANNs, however ML technically includes both of these. Another term, Deep Learning is refers only to ANNs with many layers.

AI has been used in almost every subfield of chemistry and biochemistry, more extensively in some than in others. Below I will briefly overview some examples in various subfields. These topics will be discussed in more detail later in the course.

AI in structural biology is perhaps one of the most notable successful examples of AI applied to biochemistry. While there were many incremental advances over time (which we will not discuss here), AlphaFold2 was a major advance in the use of AI to predict protein structure from protein sequence (Jumper et al). AlphaFold2 represented the first method capable of accurately predicting protein structure from sequence for many different protein families. Since AlphaFold2, there have been many additional methods reported including AlphaFold3, RoseTTAFold (Baek et al) and RoseTTAFold-All Atom (Krishna et al), ESMFold (Lin et al), Chai (Chai Discovery Team), Boltz (Wohlwend et al), and many others. While AlphaFold2 originally only had support for modeling proteins consisting of the 20 canonical amino acids, some of the more recently reported methods can model posttranslational modifications, noncanonical amino acids, ligands, metal ions, DNA, and RNA, although often with lower accuracy than proteins alone.

AI can be used to predict the molecular properties of compounds from their chemical structures, including bioactivity. Therefore, AI can be used to predict if a molecule will have a certain phenotypic activity (e.g., antimicrobial activity against a specific bacterium or antitumor activity against a certain tumor cell line) or activity against a certain protein target. The latter can be based on models that consider the structure of the protein target and the structure of the compound, otherwise high throughput screening data is generally required to train these models. An example of a model trained on phenotypic data from a high throughput screen can be found in Stokes et al and review of structure based method can be found in Sim et al. Similar models can also be used to predict other properties relevant to drug discovery such as toxicity or pharmacokinetic properties.
There are also generative methods that generate molecular structure for a desired bioactivity, either at the phenotypic or protein target level, or that will fit a target pocket in a protein. Examples include Swanson et al and Wang et al. A remaining challenge is that the generated molecules still need to be synthesized and some methods may generate structures that are challenging or impossible to synthesize.

In addition to predicting protein and peptide structure, AI can also be used to design peptides and proteins that fold into certain structures or bind certain targets. This is the reverse of the protein structure prediction problem, instead of predicting structure from sequence, sequence is predicted from structure. The desired structure can be chosen or can be designed, for example by RFDiffusion (Watson et al). The sequence can be predicted from the structure, for example by ProteinMPNN (Dauparas et al).

AI can be used in many ways in synthetic chemistry - predicting molecule reactivity, predicting if a molecule will catalyze a reaction, optimizing reaction conditions, and synthesis planning. Reaction optimization is a good application for AI because reaction condition testing can be automated and an iterative cycle between testing conditions, training a model, and testing conditions again (e.g., Angello et al). Other synthesis applications can be more difficult because they require extensive data about many chemical reactions.

In addition to predicting properties like bioactivity, AI can also be trained to predict physical properties of molecules. This includes predictions of the properties of materials from the material's structure. See Tang et al for an example.

AI is commonly used in analytical chemistry to predict various spectra from molecular structure. These include NMR, IR, and mass spectra. The reverse problem, predicting molecule from spectra, is also an area where AI models have been developed, although this is a more challenging problem. For mass spectra, especially tandem-mass spectrometry, where the molecule will be fragmented, one approach is for the model to predict which bonds in the molecule will be fragmented in order to predict the spectra. This can make predictions more accurate because the model does not have to output a precise m/z value to be correct, because that will be calculated from the predicted fragments (Nowatzky et al).

Angello, N. H.; Rathore, V.; Beker, W.; Wolos, A.; Jira, E. R.; Roszak, R.; Wu, T. C.; Schroeder, C. M.; Aspuru-Guzik, A.; Grzybowski, B. A.et al. Closed-loop optimization of general reaction conditions for heteroaryl Suzuki-Miyaura coupling. Science 2022, 378 (6618), 399.

Baek, M.; DiMaio, F.; Anishchenko, I.; Dauparas, J.; Ovchinnikov, S.; Lee, G. R.; Wang, J.; Cong, Q.; Kinch, L. N.; Schaeffer, R. D.et al. Accurate prediction of protein structures and interactions using a three-track neural network. Science 2021, 373 (6557), 871.

Chai Discovery Team; Boitreaud, J.; Dent, J.; McPartlon, M.; Meier, J.; Reis, V.; Rogozhnikov, A.; Wu, K. Chai-1: Decoding the molecular interactions of life. bioRxiv 2024, DOI:10.1101/2024.10.10.615955 10.1101/2024.10.10.615955, 2024.10.10.615955.

Dauparas, J.; Anishchenko, I.; Bennett, N.; Bai, H.; Ragotte, R. J.; Milles, L. F.; Wicky, B. I. M.; Courbet, A.; de Haas, R. J.; Bethel, N.et al. Robust deep learning-based protein sequence design using ProteinMPNN. Science 2022, 378 (6615), 49.

Jumper, J.; Evans, R.; Pritzel, A.; Green, T.; Figurnov, M.; Ronneberger, O.; Tunyasuvunakool, K.; Bates, R.; Zidek, A.; Potapenko, A.et al. Highly accurate protein structure prediction with AlphaFold. Nature 2021, 596 (7873), 583.

Krishna, R.; Wang, J.; Ahern, W.; Sturmfels, P.; Venkatesh, P.; Kalvet, I.; Lee, G. R.; Morey-Burrows, F. S.; Anishchenko, I.; Humphreys, I. R.et al. Generalized biomolecular modeling and design with RoseTTAFold All-Atom. Science 2024, 384 (6693), eadl2528.

Lin, Z.; Akin, H.; Rao, R.; Hie, B.; Zhu, Z.; Lu, W.; Smetanin, N.; Verkuil, R.; Kabeli, O.; Shmueli, Y.et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 2023, 379 (6637), 1123.

Nowatzky, Y.; Russo, F. F.; Lisec, J.; Kister, A.; Reinert, K.; Muth, T.; Benner, P. FIORA: Local neighborhood-based prediction of compound mass spectra from single fragmentation events. Nat Commun 2025, 16 (1).

Sim, J.; Kim, D.; Kim, B.; Choi, J.; Lee, J. Recent advances in AI-driven protein-ligand interaction predictions. Curr Opin Struct Biol 2025, 92, 103020.

Stokes, J. M.; Yang, K.; Swanson, K.; Jin, W.; Cubillos-Ruiz, A.; Donghia, N. M.; MacNair, C. R.; French, S.; Carfrae, L. A.; Bloom-Ackermann, Z.et al. A Deep Learning Approach to Antibiotic Discovery. Cell 2020, 181 (2), 475.

Swanson, K.; Liu, G.; Catacutan, D. B.; Arnold, A.; Zou, J.; Stokes, J. M. Generative AI for designing and validating easily synthesizable and structurally novel antibiotics. Nature Machine Intelligence 2024, 6 (3), 338.

Tang, Y. H.; Xu, W. B.; Cao, J.; Gao, W. L.; Farrell, S.; Erichson, B.; Mahoney, M. W.; Nonaka, A.; Yao, Z. J. A multimodal large language model for materials science. Nature Machine Intelligence 2026, 8 (4).

Wang, J.; Zhang, D. Y.; Budakoti, S.; Dokholyan, N. V. A diffusion-based framework for designing molecules in flexible protein pockets. Sci Adv 2026, 12 (15), eaeb7045.

Watson, J. L.; Juergens, D.; Bennett, N. R.; Trippe, B. L.; Yim, J.; Eisenach, H. E.; Ahern, W.; Borst, A. J.; Ragotte, R. J.; Milles, L. F.et al. De novo design of protein structure and function with RFdiffusion. Nature 2023, 620 (7976), 1089.

Wohlwend, J.; Corso, G.; Passaro, S.; Getz, N.; Reveiz, M.; Leidal, K.; Swiderski, W.; Atkinson, L.; Portnoi, T.; Chinn, I.et al. Boltz-1 Democratizing Biomolecular Interaction Modeling. bioRxiv 2025, DOI:10.1101/2024.11.19.624167 10.1101/2024.11.19.624167, 2024.11.19.624167.