DUE: 11:59 PM TUESDAY, DECEMBER 8
Design and implement a small research project that aims to answer a question involving macromolecular and/or chemical structure. We recommend working in groups of two. You may work in a group of three, but this will result in an expectation that the project will involve 50% more effort. You may work solo, but this will not result in any difference in the amount of effort expected relative to a two-person project.
Your final deliverable will be a research poster, due 11:59pm Tuesday, December 8. All posters must be printed in time for a poster session from noon-2:30pm on Wednesday, December 9. Posters must be portrait orientation and no larger than 36" x 42". Background should be white (to avoid excessive use of ink).
Posters must succinctly yet comprehensively explain the research project, including providing necessary background and a clear description of the research question, methods, and results. You also have the option of providing a maximum of 5 pages of supplementary material (not including references). This supplemental material is not required, but as you are strongly encouraged to create a visually engaging poster that is not a wall of text or bogged down in extraneous detail, the supplement provides a means to provide information about your project that you would like to be considered during grading but that does not easily fit on a small poster (e.g., extended background, algorithms, code listings, proofs, additional supporting but not critical data analysis). The supplemental material is not expected to be a standalone research paper (i.e., you can assume the reader is already familiar with the poster).
You can find a very basic poster template here. Feel free to make any changes, but stick to the white background requirement.
Project Ideas
You should define and answer a research question. This may involve using "off-the-shelf" tools and methods discussed in class or may involve proposing improvements to methods or a rigorous evaluation of methods. Projects should be appropriately scoped for the time available. For example, you will not be able to train a structure prediction model from scratch (Boltz-1 took a month on a cluster of H100s), but you could likely fine-tune a pre-trained model or use a pre-trained model's embeddings in a downstream task. Some ideas:
- Drug discovery: Identify putative novel modulators of therapeutically relevant targets. For a challenge, consider an undruggable target. A drug discovery project needs to be more involved than Assignment 7. For instance, performing modeling to identify a putative allosteric site and then running simulations with your proposed ligand to evaluate efficacy. Alternatively, you can run free energy calculations on your hits (you may even start with your NRAS hits). Ideally you will have a positive control (known binder, such as GDP/GTP for NRAS) in a free energy calculation project.
- Mechanisms: Perform simulations to compare the dynamics and functions of different proteins. For example, using weighted ensemble to perform binding simulations of an antibody with two different antigens (or two different antibodies with the same antigen).
- Method evaluation: How well do state-of-the-art structure prediction models correlate to function? Evaluate Boltz and/or AlphaFold (or others) on a deep mutational scan dataset to see the extent to which they can provide insight into the effect of point mutations. How do various models differ in their internal representations of proteins? How do the properties of the ensemble of structures generated by diffusion models compare to the true Boltzmann distribution? Do the per-residue embeddings learned in these models correlate to any functional readouts (e.g., allosteric networks)?
- Method improvements: Can data-driven ML models be guided by physics to produce better results (e.g., including force field calculations during diffusion)? Can protein foundation models be extended to predict functional readouts or mutational effects (e.g., allosteric signaling)? Can large models be successfully knowledge distilled? Can LoRA be successfully applied to these models?
Project Proposal
Add a single slide to this presentation by 11:59am Wednesday, November 11 (that is, before start of class). Your slide should provide the question your project seeks to answer and a high-level description of your proposed methods. Include links to any relevant papers. Figure or preliminary data are not needed. You are strongly encouraged to submit your proposal earlier. You must submit a slide, even if you get a verbal approval. As soon as your proposal is accepted, you may start working on your project.