Machine Learning Ensemble Directed Engineering of Genetically Encoded Fluorescent Calcium Indicators

Abstract

Real-time monitoring of biological activity can be achieved through the use of genetically encoded fluorescent indicators (GEFIs). GEFIs are protein-based sensing tools whose biophysical characteristics can be engineered to meet experimental needs. However, GEFIs are inherently complex proteins with multiple dynamic states, rendering optimization one of the most challenging problems in protein engineering. Most GEFIs are engineered through trial-and-error mutagenesis, which is time and resource-intensive and often relies on empirical knowledge for each GEFI. We applied an alternative approach using machine learning to efficiently predict the outcomes of sensor mutagenesis by analyzing established libraries that link sensor sequences to functions. Using the GCaMP calcium indicator as a scaffold, we developed an ensemble of three regression models trained on experimentally derived GCaMP mutation libraries. We used the trained ensemble to perform an in silico functional screen on a library of 1423 novel, untested GCaMP variants. The mutations were predicted to significantly alter the fluorescent response, and off-rate kinetics were advanced for verification in vitro. We found that the ensembles predictions of novel variants biophysical characteristics closely replicated what we observed of the variants in vitro. As a result, we identified the novel ensemble-derived GCaMP (eGCaMP) variants, eGCaMP and eGCaMP+, that achieve both faster kinetics and larger fluorescent responses upon stimulation than previously published fast variants. Furthermore, we identified a combinatorial mutation with extraordinary dynamic range, eGCaMP2+, that outperforms the tested 6th, 7th, and 8th generation GCaMPs. These findings demonstrate the value of machine learning as a tool to facilitate the efficient prescreening of mutants for functional characteristics. By leveraging the learning capabilities of our ensemble, we were able to accelerate the identification of promising mutations and reduce the experimental burden associated with screening an entire library. Machine learning tools such as this have the potential to complement emerging high-throughput screening methodologies that generate massive datasets, which can be tedious to analyze manually. Overall, these findings have significant implications for developing new GEFIs and other protein-based tools, demonstrating the power of machine learning as an asset in protein engineering.

Biophysics-based protein language models for protein engineering

“…Machine learning-guided protein engineering is often data-limited due to experimental throughput constraints, with datasets sometimes containing as few as tens to hundreds of sequence-function examples [31, 68–73]. We demonstrated METL’s performance in realistic protein engineering settings with limited data (low-N) and extrapolation.…”