Enhanced Sampling\nof Biomolecular Slow Conformational\nTransitions Using Adaptive Sampling and Machine Learning

Biomolecular simulations often suffer from the “time\nscale\nproblem”, hindering the study of rare events occurring over\nextended time scales. Enhanced sampling techniques aim to alleviate\nthis issue by accelerating conformational transitions, yet they typically\nnecessitate well-defined collective variables (CVs), posing a significant\nchallenge. Machine learning offers promising solutions but typically\nrequires rich training data encompassing the entire free energy surface\n(FES). In this work, we introduce an automated iterative pipeline\ndesigned to mitigate these limitations. Our protocol first utilizes\na CV-free count-based adaptive sampling method to generate a data\nset rich in rare events. From this data set, slow modes are identified\nusing Koopman-reweighted time-lagged independent component analysis\n(KTICA), which are subsequently leveraged by on-the-fly probability\nenhanced sampling (OPES) to efficiently explore the FES. The effectiveness\nof our pipeline is demonstrated and further compared with the common\nMarkov State Model (MSM) approach on two model systems with increasing\ncomplexity: alanine dipeptide (Ala2) and deca-alanine (Ala10), underscoring\nits applicability across diverse biomolecular simulations.