DeepMind vs. Academia: How good are the best AI models in discriminating similar sequences of a given protein?- Group 1

2021-2022 Fall
Faculty Department of Project Supervisor: 
Faculty of Engineering and Natural Sciences
Number of Students: 

For too many years, the protein folding problem, i.e., determining the three-dimensional structure of a protein from a knowledge of the sequence alone, has been a conundrum for researchers. This is a very important problem, as knowing the shape of a protein is essential for solving a whole range of biological problems such as finding the mechanism of action of mutants arising in diseases, and designing drugs to inhibit functions of proteins, to name just a couple. While most of us thought the solution would come from developing a physics-based understanding of how atoms in the molecule interact, last year DeepMind developed an advanced protein structure prediction artificial intelligence model called AlphaFold that performed way better than any previously developed approach[1]. Meanwhile, academics who had been working on the problem on the Robetta server came up with a similar AI model soon after[2], and the two are now open for to the use of researchers all around the world. Both developments are hailed as breakthroughs both in the biological and the computational sciences research communities [3,4].  In this project, we will take up hard cases where the sequence of a protein whose structure is known will be varied to find out the level at which the limits of these AI models may be pushed.
1. 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, 596:583.
2. Baek M, DiMaio F, Anishchenko I, Dauparas J, Ovchinnikov S, Lee GR, Wang J, Cong Q, Kinch LN, Schaeffer RD, et al.: Accurate prediction of protein structures and interactions using a three-track neural network. Science 2021, 373:871.
3. AlQuraishi M: Protein-structure prediction revolutionized. Nature 2021, 596:487.
4. Researchers unveil ‘phenomenal' new AI for predicting protein structures on World Wide Web URL:

Related Areas of Project: 
Computer Science and Engineering
Molecular Biology, Genetics and Bioengineering
Materials Science ve Nano Engineering

About Project Supervisors

Canan Atilgan
personal web page:
research group web page:
Please get ready to talk to me before signing up to this project. You may schedule an appointment via email: I will be inquiring about your interests on this particular topic so as to find the best matching students for the project. Ideally we would have one CS and one BIO major working together and exchanging ideas. However, MAT students may also be a good fit. An acute interest in coding is essential, although proficiency is not required (translates as: we will teach you if you are interested).
Tandaç Furkan Güçlü
Faculty of Engineering and Natural Science
Postdoctoral Researcher