But it may well be semi-supervised.
For some time now I have thought that building a latent representation of protein sequence space is a really good idea, both because we have far more sequences than any form of labelled data, and because, once built, such a representation can inform a broad range of downstream tasks. This is why I jumped at the opportunity last year when Surge Biswas, from the Church Lab, approached me about collaborating on exactly such a project. Last week we posted a preprint on bioRxiv describing this effort. It was led by Ethan Alley, Grigory Khimulya, and Surge. All I did was to enthusiastically cheer them on, and so the bulk of the credit goes to them and George Church for his mentorship.
I just came back from CASP13, the biennial assessment of protein structure prediction methods (I previously blogged about CASP10.) I participated in a panel on deep learning methods in protein structure prediction, as well as a predictor (more on that later.) If you keep tabs on science news, you may have heard that DeepMind’s debut went rather well. So well in fact that not only did they take first place, but put a comfortable distance between them and the second place predictor (the Zhang group) in the free modeling (FM) category, which focuses on modeling novel protein folds. Is the news real or overhyped? What is AlphaFold’s key methodological advance, and does it represent a fundamentally new approach? Is DeepMind forthcoming in sharing the details? And what was the community’s reaction? I will summarize my thoughts on these questions and more below. At the end I will also briefly discuss how RGNs, my end-to-end differentiable model for structure prediction, did on CASP13.
For over a decade now I have been working, essentially off the grid, on protein folding. I started thinking about the problem during my undergraduate years and actively working on it from the very beginning of grad school. For about four years, during the late 2000s, I pursued a radically different approach (to what was current then and now) based on ideas from Bayesian nonparametrics. Despite spending a significant fraction of my Ph.D. time on the problem, I made no publishable progress, and ultimately abandoned the approach. When deep learning began to make noise in the machine learning community around 2010, I started thinking about reformulating the core hypothesis underlying my Bayesian nonparametrics approach in a manner that can be cast as end-to-end differentiable, to utilize the emerging machinery of deep learning. Today I am finally ready to start talking about this long journey, beginning with a preprint that went live on bioRxiv yesterday.
There has been a lot of renewed interest lately in neural networks (NNs) due to their popularity as a model for deep learning architectures (there are non-NN based deep learning approaches based on sum-products networks and support vector machines with deep kernels, among others). Perhaps due to their loose analogy with biological brains, the behavior of neural networks has acquired an almost mystical status. This is compounded by the fact that theoretical analysis of multilayer perceptrons (one of the most common architectures) remains very limited, although the situation is gradually improving. To gain an intuitive understanding of what a learning algorithm does, I usually like to think about its representational power, as this provides insight into what can, if not necessarily what does, happen inside the algorithm to solve a given problem. I will do this here for the case of multilayer perceptrons. By the end of this informal discussion I hope to provide an intuitive picture of the surprisingly simple representations that NNs encode.
It is tempting to assume that with the appropriate choice of weights for the edges connecting the second and third layers of the NN discussed in this post, it would be possible to create classifiers that output over any composite region defined by unions and intersections of the 7 regions shown below.
Continue reading ∞