| Instructor | Ellen Zhong (zhonge@princeton.edu), CS 314 |
| TA | Ziyu Xiong (ziyux@princeton.edu) |
| Time | Wednesdays, 2:55-4:55pm |
| Location | Bendheim House 103 |
| Office hours | Tuesdays 3:00-4:00pm, CS 314, or by appointment |
This course explores how modern machine learning methods are used in the natural sciences, with a focus on molecular and biological applications. We will study the fundamental principles and design of AI systems and machine learning models across a variety of scientific data modalities, including sequences, images, 3D structures, and dynamical data, and examine their use in prediction, inference, generation, and design tasks. Application areas include biological language models, protein structure prediction, de novo protein design, 3D reconstruction in cryo-EM, molecular simulation, small molecule drug discovery, and AI agents for scientific discovery. Through lectures, paper discussions, and a final project, students will learn both machine learning foundations and emerging research directions in AI for the natural sciences.
Course communication. We will use a Slack team for most communications this semester. You will be added to the Slack team after the first week. If you join the class late, email Ellen and Ziyu, and we'll add you. Once you're on Slack, we prefer Slack messages over emails for all logistical questions. We also encourage students to use Slack for discussions related to lecture content and projects.
Presentation scheduling. All students should complete a short Google Form on presentation preferences and scheduling before the second lecture (Sept. 9).
This course combines traditional lectures with presentations and discussions on the primary literature. AI for the natural sciences is a rapidly evolving field, and many of the most important ideas are best learned by reading and critically evaluating contemporary research papers. There are two main components of this course:
Lectures will introduce the core concepts and technical foundations for the week's topic, focusing on the fundamental ideas underlying modern AI methods for the natural sciences and providing the background needed to understand and critically evaluate current research.
Student-led paper presentations. Each week, there will be one or two required research papers related to the week's lecture topic. Depending on enrollment, 1-2 students will work together and present the assigned paper to the class.
The goal of the presentation is not simply to summarize the paper, but to teach it to the class and lead a thoughtful discussion. Presenters should explain the motivation behind the work, provide sufficient technical background, critically evaluate the methodology and results, and place the paper in the broader context of the field. The presentation schedule and paper assignments will be determined during the first weeks of the semester.
Pre-lecture feedback. Presenters are required to meet with the instructor at least one day prior to their presentation to review the organization and content of their slides. Draft slides should be submitted by 11:59 PM on the day before the scheduled meeting.
A suggested presentation outline is:
Participation. All students are expected to read the assigned paper(s) before class, fill out a Google form containing pre-lecture questions, and actively participate in discussions.
Post-lecture feedback. All non-presenters will be responsible for providing feedback to the presenters via a Google form. A few bullet points of constructive feedback on what they liked, what was clear, and suggestions for improvement will be due by the end of the day on Wednesday. Feedback is a gift!
(Tentative) For some weeks, we may instead have guest speakers come in and discuss their work for the first hour of the class (or at an alternate time, depending on scheduling), which will be followed by a discussion.
The course includes a semester-long research project that provides students with the opportunity to explore a machine learning problem in the natural sciences in greater depth. Students may work individually or in groups of up to three.
The project is structured around developing a research proposal or preliminary research study in the style of a machine learning workshop paper. The goal is to identify an interesting scientific problem, motivate the approach, connect it to existing literature, and evaluate the feasibility of the proposed methodology through analysis and/or preliminary experiments.
Throughout the semester, students will submit a short project proposal, receive feedback, and present their work during the final weeks of the course. The project is intended to encourage students to think creatively about new research directions while gaining hands-on experience formulating and communicating research ideas at the intersection of machine learning and the natural sciences.
The breakdown of the final grade is:
Comment on participation: Note that a major part of the grade is attending class each week and engaging in discussions. It is important to monitor your own participation. Participation in discussions includes posing and answering questions, explaining background material or literature, providing constructive feedback on papers, and brainstorming future avenues of research.
Attendance is required in all classes. If possible, the instructor must be notified in advance of any extenuating circumstances that lead to missed attendance, and arrangements need to be made to make up for missed work.
Students should have prior exposure to introductory machine learning concepts and deep learning architectures. No prior knowledge of biology and chemistry is required, however, students should be interested in learning and gaining domain expertise. Students should expect to develop a sufficient understanding of each application area to evaluate new developments in this fast-paced, dynamic field.
A tentative schedule of weekly topics is presented below.
| Week | Dates | Topic | Readings | Presenters | Pre-lecture questions |
|---|---|---|---|---|---|
| 1 | Sept 2 | Introduction: AI for Science | Ellen Zhong [Slides] | ||
| 2 | Sept 9 | Breakthroughs in AI for Science: Where did they come from? |
1. Jumper, J., Evans, R., Pritzel, A. et al. Highly accurate protein structure
prediction with Alphafold. Nature 2021.
Optional readings: 2. Abramson, J., Adler, J., Dunger, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 2024. 3. Tunyasuvunakool, K., Adler, J., Wu, Z. et al. Highly accurate protein structure prediction for the human proteome. Nature 2021. 4. Blog post from CASP14: AlphaFold2 @ CASP14: "It feels like one's child has left home." |
Ellen Zhong [Slides] | Link |
| 3 | Sept 16 | Intro deep learning and generative modeling |
1. Kingma, D.P. & Welling, M. Auto-Encoding Variational
Bayes. ICLR 2014.
2. Zhong et al. Reconstructing continuous distributions of 3D protein structure from cryo-EM images. ICLR 2020. |
Ellen Zhong [Slides], Jack Nugent, Sol Park [Slides-VAE], Philipp Janke, Ariana Vlad [Slides-cryoDRGN] | Link |
| 4 | Sept 23 | Transformers and scaling laws |
1. Vaswani, A., Shazeer, N., Parmar, N. et al. Attention is
all you need. NeurIPS 2017.
2. Language Modeling Materializes a World Model of Protein Biology. bioRxiv 2026. Optional readings: 3. Kaplan, J., McCandlish, S., Henighan, T. et al. Scaling Laws for Neural Language Models. arXiv 2020. 4. Hoffmann, J., Borgeaud, S., Mensch, A. et al. Training Compute-Optimal Large Language Models. NeurIPS 2022. |
Zeming Lin (Guest speaker, ESM/Biohub) | Link |
| 5 | Sept 30 | Diffusion & flow matching models |
1. Ho, J., Jain, A. & Abbeel, P. Denoising Diffusion
Probabilistic Models. NeurIPS 2020.
2. Ingraham, J.B. et al. Illuminating protein space with a programmable generative model. Nature 2023. Optional readings: 3. Song, Y., Sohl-Dickstein, J., Kingma, D.P. et al. Score-Based Generative Modeling through Stochastic Differential Equations. ICLR 2021. 4. Lipman, Y., Chen, R.T.Q., Ben-Hamu, H. et al. Flow Matching for Generative Modeling. ICLR 2023. 5. Albergo, M.S. & Vanden-Eijnden, E. Building Normalizing Flows with Stochastic Interpolants. ICLR 2023. 6. Albergo, M.S., Boffi, N.M. & Vanden-Eijnden, E. Stochastic Interpolants: A Unifying Framework for Flows and Diffusions. JMLR 2025. |
Ellen Zhong, Binxu Li, Haoyi Duan, Yury Polyachenko, Jiashi Yin | |
| 6 | Oct 7 | Inverse problems; generative models for scientific inference | Ellen Zhong, Sreemanti Dey, Rishi Kannan, Karan Tandon, Ishan Saha | ||
| 7 | Oct 14 | AI agents for science | Daniel Lima Braga, Sanketh Kamath, Kelsey Fu, Tao Zhong, Linrong Cai, Clemens Sageder | ||
| 8 | Oct 21 | Fall break (no class) | |||
| 9 | Oct 28 | AI for chemistry, geometric deep learning | Levi Rauchwerger, Aryan Saha, Edward Wu, Jina Kim, Tom Yin | ||
| 10 | Nov 4 | AI for physics | Ram Narayanan, Alexander Kreibich, Sanjay Kumar Keshava, Dongyeob Kim | ||
| 11 | Nov 11 | Guest lecture: Bowen Jing | Bowen Jing (guest speaker) | ||
| 12 | Nov 18 | Interpreting scientific foundation models | Bruce Shen, Tyler King, Taiming Lu, Dongzhe Zheng | ||
| 13 | Nov 25 | Thanksgiving break (no class) | |||
| 14 | Nov 30 - Dec 2 (both days, tentative) |
Final project presentations | |||
| 15 | Dec 7 (Monday) | Final project presentations |