6.1 Integrating MSM and ML to elucidate the emergence of
function in complex systems
We are riding the wave of a paradigm shift in the development of MSM
methods due to rapid development and changes in HPC infrastructure (see
Figure 2) and advances in ML methods. Thus, MSM and HPC have emerged as
essential tools for modeling complex problems at the microscopic scales
with a focus on leveraging the structured and embedded physical laws to
gain a mechanism-based understanding. This success notwithstanding, the
design of new MSM algorithms in coupling different scales, data
utilization, and their implementation on HPC is becoming increasingly
cumbersome in the face of heterogeneous data availability and rapidly
evolving HPC architectures and platforms. On the other hand, while
purely data-driven models of molecular and cellular systems spawned by
the techniques of data science [132-134], and in particular, ML
methods including deep learning methods [135-137], are easy to train
and implement, the underlying model manifests as a black-box. This
general approach taken by the ML community is well suited for
classification, learning, and regression problems, but suffers from
limitations in interpretability and explainability, especially when
mechanism-based understanding is a primary goal. There lies a vast
potential in combining MSM, HPC, and ML methods with their complementary
strengths [4]. MSM models are routinely coupled together by
appropriately propagating information across scales (see section 5),
while the ever-increasing advances in hardware capabilities and
high-performance software implementations allows us to study
increasingly more complex phenomena at a higher fidelity and higher
resolution. While much of the discussion thus far has been focused on
MSM and HPC methods, the progress and potential in integrating MSM and
ML are discussed below and represent the forefront of emerging MSM
research, in which we discuss a few emerging integrative approaches to
combine ML and MSM.