Paradigm Shift: The Promise of Deep Learning in Molecular Systems Engineering and Design

Abstract

The application of deep learning to a diverse array of research problems has accelerated progress across many fields, bringing conventional paradigms to a new intelligent era. Just as the roles of instrumentation in the old chemical revolutions, we reinforce the necessity for integrating deep learning in molecular systems engineering and design as a transformative catalyst towards the next chemical revolution. To meet such research needs, we summarize advances and progress across several key elements of molecular systems: molecular representation, property estimation, representation learning, and synthesis planning. We further spotlight recent advances and promising directions for several deep learning architectures, methods, and optimization platforms. Our perspective is of interest to both computational and experimental researchers as it aims to chart a path forward for cross-disciplinary collaborations on synthesizing knowledge from available chemical data and guiding experimental efforts.

Next generation pure component property estimation models: With and without machine learning techniques

“…Recently, ML has emerged as an alternative modeling method that can play a transformative role across many disciplines 26 . This is performed by uncovering intricate relationships in high‐dimensional spaces and allows a systematic and thorough search of the hypothesis space for a suitable model‐function 27,28 .…”

“…Recently, ML has emerged as an alternative modeling method that can play a transformative role across many disciplines. 26 This is performed by uncovering intricate relationships in high-dimensional spaces and allows a systematic and thorough search of the hypothesis space for a suitable model-function. 27,28 In the realm of molecular systems, ML has been applied to predict pure component properties, 14 reduce the computational cost associated with quantum mechanics/molecular mechanics calculations, 29 generate lead molecules while efficiently exploring the chemical space, 30 and even create a novel QSPR methods, 31 and predict reaction outcomes and retrosynthesis.…”

Versatile Deep Learning Pipeline for Transferable Chemical Data Extraction

“…1−3 Recently, the confluence of "big data" from chemical literature and deep learning has led to significant advancements in synthesis planning, 4,5 protein folding, 6,7 molecular and drug design, 8−10 and conformer generation, 11,12 challenging years of traditional theorization and experimentation. 13 However, advancements in emerging areas, including battery materials, 14 biomaterials, 15 and microplastics, 16 are often hindered by the lack of structured data sets that can facilitate rapid discovery. 17,18 Though invaluable, the manual curation of these relational data sets is expensive, labor-intensive, and error-prone, often requiring domain-specific expertise.…”

“…Chemical knowledge, predominantly present in the form of unstructured texts, is spread across a vast and continually growing body of literature. The extensive literature contains a plethora of relational chemical data that holds the potential to drive breakthroughs across multiple domains. − Recently, the confluence of “big data” from chemical literature and deep learning has led to significant advancements in synthesis planning, , protein folding, , molecular and drug design, − and conformer generation, , challenging years of traditional theorization and experimentation . However, advancements in emerging areas, including battery materials, biomaterials, and microplastics, are often hindered by the lack of structured data sets that can facilitate rapid discovery. , Though invaluable, the manual curation of these relational data sets is expensive, labor-intensive, and error-prone, often requiring domain-specific expertise .…”

End-to-end learning with interpretation on electrohysterography data to predict preterm birth

“…While this has led to valuable information, designing handcrafted features from EHG data does lead to loss of information. In recent years, there has been a shift from feature engineering to end-to-end learning; namely Deep Learning (DL) [26], which has been applied in many medically related research areas, including clinical imaging [27]. Nevertheless, to the best of our knowledge, these models have not been used to predict preterm birth using EHG data as time series.…”