we had 3 papers to read for today, but only discussed one “Predicting the effects of non-coding variants..” (2015) Zhou and Troyanskaya. It had impressive results but was frustrating to read since details of both data and model were not described.

Wednesday 9AM we will discuss “END-TO-END DIFFERENTIABLE LEARNING OF PROTEIN STRUCTURE” (2018) by Mohammed AlQuraishi who has solved the protein folding problem. And “Accurate De Novo Prediction of Protein Contact Map by Ultra-Deep Learning Model” (2017)

Since early May we have met Monday Wednesday and Friday 9AM reading books and papers relevant to DL. We started with reading/discussing 700 pages in Goodfellow (2016) et al : Deep Learning and Giron (2017) Hands-On using TensorFlow. after that we switch to papers from different application areas and have taken papers on Games (GO, Backgammon, Atari), Biosciences (Baldi, 2018) and are now in the middle of Genotype–>Phenotype mapping. We have 4 weeks to go and hope to cover protein structure, chemoinformatics, finance and a few deeper methodological papers from people in the Department. Any can join and come with suggestions about what to read.

I hope I am wrong – I worked hard for this not to be the case, but maybe not hard enough. I am giving a graduate lecture – 4th in 18 months on “Algorithms, Combinatorics in ChemoInformatics”.

Title: Combinatorics and Algorithms in ChemoInformatics Venue: May 17th 3.30 PM Department of Statistics, LG.04

Summary: Chemoinformatics is central to Drug Development and Design. In this lecture, we will go through key algorithms and combinatorics related to Chemoinformatics. Such algorithms are graph isomorphism, subgraph isomorphism, maximal common subgraphs and double pushout graph grammars. Combinatorics include generating functions for counting/enumerating special classes of molecules starting with alkanes, Polya-counting/enumerating molecules with symmetries, recursive enumeration of molecular graphs. We will also mention calculation of synthetic pathways, prediction of reactions and catalysis, exploration of chemical space and the potential for the use of Deep Learning. The talk attempts to survey these techniques in a way that should be useful for users that normally don’t venture into these techniques but maybe use chemoinformatics tools. 90 minutes – 30 slides:

https://tinyurl.com/algochemo

We are done with Roemer (1996): “Theories of Distributive Justice”.  It was hard to digest to say the least and we might still have a wrap-up dinner at University College, if I find an expert who are willing to discuss with the 4 readers.

We are real happy to move on and somehow we found the above topic appealing.  I have avoided reading on this as I feel the Brain extremely complex, textbooks on Neuroscience are often huge, I am sceptical about the contribution about philosophy to the kind of knowledge I strive for.  I personally feel absolutely fine with a life without consciousness and have repeatedly recommended it to others.

But it is one of the BIG QUESTIONS and the ability to simulate the brain grows, so it would be real exciting to scratch the surface of this topic.   I have also included 4 lectures in my course Topics in Computational Biology in 2019, so I better get started reading the background literature and for once this fits right in.

I emailed 5 Professors of Neuroscience I knew, for recommendations and so far this book got unambigous praise, but I hope for a bit more feedback and advice on supplementary readings.  We are real careful with the books we read and it is not easy to get through the needle’s eye.  It has to be hard to read so it needs thorough discussion for each chapter.  In Roemer we at times used 30 minutes per page…..

But if I don’t find a better book by Monday morning, we will start Friday May 4th 6.30AM UK time per skype to discuss Stan Dehaene (2014): “Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts”.  We might supplement this with technical literature if need be.

Professor Julián Echave will be giving an informal talk on Tuesday 10 April at 11am in the Small Lecture Theatre at the Department of Statistics.

Title: Protein evolutionary divergence is not random

Abstract

A simple comparison of homologous proteins shows clear patterns of differential conservation/variation at the levels of amino-acid sequence, 3D structure, and protein motions. For instance, the rate of sequence evolution varies among sites; protein structures diverge more at some sites than others, and some protein vibrations are more variable than others. The default explanation of evolutionary patterns is the rather fuzzy concept of “functional importance”: the underlying assumption is that any extra conservation/variability is due to natural selection. However, while selection does indeed shape sequence divergence, the patterns of divergence of structure and motion are mostly shaped by the physics of the response of proteins to random mutations.