What are some of the interesting projects you have worked on in the field of Computational Biology? originally appeared on Quora - the knowledge sharing network where compelling questions are answered by people with unique insights.
Answer by Sebastian Raschka, Author of Python Machine Learning. Computational Biology PhD candidate at MSU, on Quora.
Most of my other projects are concerned with large-scale virtual screening applications: In collaboration with labs that do experimental biology, we develop and apply methods to predict candidate molecules that either inhibit (or activate, depending on the project) an individual protein, in absence or presence of a protein crystal structure. The interesting part about it is the interplay between predictions and feedback: I get to make predictions and (at some point), I get the experimental results to see whether I was right or wrong and to analyze why certain predictions worked better than others. Another exciting challenge in such projects is that one has to find ways to make this all computationally feasible -- if you have 15 million molecules, selecting 100 candidate molecules for experimental testing is a bit like searching for the needle in the haystack. Usually, it comes down to formulating specific hypotheses upfront as "filtering" steps since a brute-force docking, which computationally not feasible (since we also have time constraints). My projects require a certain amount of creativity and technical skills to put the ideas into action, but eventually, the approach (the hypotheses) also have to make sense to our collaborators (and the funding agencies).
Besides these applications of virtual screening (some of them are completed and I am currently in the process of writing up the papers and releasing the toolkit(s)), I also work on more general concepts regarding protein-ligand interactions. We recently discovered an interesting phenomenon related to protein-ligand interactions, and we are looking at more data points to make sure that this is not just an artifact. Another project I just wrapped up this summer would be about computing the local rigidity of protein-ligand binding pockets to predict the near-native protein-ligand binding mode (Detecting the native ligand orientation by interfacial rigidity: SiteInterlock). I thought it was an interesting idea since near-native binding modes are typically predicted via a sum of different energy terms. Using rigidity theory, counting the degrees of freedom in the structures is more about the cooperativity of interactions rather than their additive sum. Or in other words, if a particular non-covalent interaction does not remove additional degrees of freedom from the complex (if the complex is already rigid), it is not "counted" towards the interaction score. In practice, using the local rigidity of a protein-ligand complex seems to perform as well as other empirical or knowledge-based scoring functions, and in addition to being a "standalone" scoring function, I see it as an interesting new "signal" or "feature" for ensemble scoring.
This question originally appeared on Quora. - the knowledge sharing network where compelling questions are answered by people with unique insights. You can follow Quora on Twitter, Facebook, and Google+.