Anna Visibelli (University of Siena)
Jan 15, 2020 – 11:00 AM
DIISM, Artificial Intelligence laboratory (room 201), Siena SI
Microbial rhodopsins, light-dependent ion-transporting membrane proteins, are widely used in optogenetics, due to their low phototoxicity and high tissue penetration. To widen possible optogenetic applications (e.g., diverse optical control of neural activity) novel rhodopsins, exhibiting specific photochemical properties, are necessary. However, the experimental expression and testing of hundreds, if not thousands, of rhodopsins and their single and multiple mutants is not feasible, in terms of costs and time. Alternatively, computational approaches can help to narrow down such numbers to few tens of likely successful candidates. In this regard, the Automatic Rhodopsin protocol (a-ARM) is a suitable computational tool, not only for the generation of the mutated structures, but also for the production of hybrid quantum mechanics/molecular mechanics (QM/MM) models useful for the prediction of photochemical properties (i.e. absorption and emission wavelengths). However, a-ARM still requires ca. 36 hours in our cluster facility to produce one QM/MM model, which amounts to, ideally, ca. 23 models per week, given the number of currently available computing nodes. Recently, Karasuyama et al. reported a machine-learning-based (ML-based), data-driven approach, which uses a simple relationship between amino-acid sequences and absorption wavelengths. In this work, we decided to complement a-ARM with a new ML-based pre-screening process, and integrate them into a new protocol, called a-ARM-ML.