BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//ZContent.net//ZapCalLib 1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
SUMMARY:Advancing Solubility Prediction Through Machine Learning
LOCATION:Chemistry A101
TZID:America/Denver
DTSTART:20250428T160000
UID:2026-05-05-09-32-30@natsci.colostate.edu
DTSTAMP:20260505T093230
Description:About the Seminar: \n\nSolubility is a fundamental chemical pr
 operty with wide-ranging applications including reaction optimization\, wa
 ste recycling\, and manufacturing. As measuring solubility can be time or 
 resource-intensive\, predicting solubility through computational methods h
 as received significant attention in recent work. In particular\, solubili
 ty prediction through machine learning (ML) has been heavily studied due t
 o its speed and accessibility advantages over alternative methods using qu
 antum mechanical or semi-empirical formulations. In this talk\, we discuss
  two recent advances in solubility prediction for polymers and small molec
 ules respectively. We first discuss our recent work to predict polymer sol
 ubility in single solvents\, which is of interest for applications in plas
 tic recycling and polymer design. We found that simple tree-based models w
 ith low-dimensional features can achieve over 80% prediction accuracy on h
 omopolymer and copolymer solubility\, and that these predictions can be ra
 tionalized using explainable AI methods such as Shapley Additive Explanati
 ons (SHAP). Following our discussion of polymer solubility prediction\, we
  next examine ML predictions of small molecule solubility in multiple solv
 ents (multicomponent solubility). In comparison to single solvent solubili
 ty\, multicomponent solubility has increased complexity but allows for gre
 ater control over solute separation and processing\, leading to uses in bi
 omass upgrading and recycling. To accelerate these applications\, we curat
 ed a new multicomponent solubility database (MixSolDB) which we used to tr
 ain two graph neural network (GNN) models to predict solute solubility in 
 up to three solvents. We find that our novel subgraph architecture for sol
 ubility prediction outperforms the more common concatenation architecture\
 , achieving a mean absolute error (MAE) of 0.67 kcal/mol on ΔGsolv predic
 tion. In summary\, we demonstrate that ML-based predictions of solubility 
 are chemically accurate while remaining useful for sustainable application
 s. 4:00 pm
END:VEVENT
END:VCALENDAR