Clustering protein structures with Hadoop

Machine learning is a widely used technique in structural biology, since the analysis of large conformational ensembles originated from single protein structures (e.g. derived from NMR experiments or molecular dynamics simulations) can be approached by partitioning the original dataset into sensible subsets, revealing important structural and dynamics behaviours. Clustering is a good unsupervised approach for dealing with these ensembles of structures, in order to identify stable conformations and driving characteristics shared by the different structures. A common problem of the applications that implement protein clustering is the scalability of the performance, in particular concerning the data load into memory. In this work we show how it is possible to improve the parallel performance of the GROMOS clustering algorithm by using Hadoop. The preliminary results show the validity of this approach, providing a hint for future development in this field.