The Kim Group had a blast attending the ACS conference in San Francisco! Every member showcased their latest work, with Yeonjoon shining brightly as he presented his research. The rest of us had engaging discussions around our poster presentations. Big cheers to our fellow CSU groups who joined the event – it was great seeing our university so well-represented and vibrant!
Yeonjoon presented the work regarding the recent development of the graph neural network (GNN) model for predicting the cetane number of multicomponent mixtures. This GNN copes with the increasing complexity of various mixture compositions by introducing inductive biases based on the underlying equations of thermodynamics of mixing, achieving reliable accuracies for up to 13-component mixtures.
At the ACS conference Raúl presented his work regarding his work on the usage of Machine learning models to understand complex phenomena such as the Catalytic Fast Pyrolysis process on different biomass feedstocks. The work ranges from the application of ML models to elucidate the effects of different catalysts to the prediction of how a specific catalyst would affect the spectra of uncatalyzed Fast Pyrolysis vapors.
At ACS Fall 23, Chris presented his work on predicting polymer solubility using machine learning, with a focus on model explainability and in-house experimental validation. This research examined a wide range of descriptors and architectures to predict polymer solubility, including graph-based and descriptor-based models.
Sabari presented the preliminary results of his protein solubility prediction model. This work introduces a novel neural network architecture that incorporates geometric and topological information about protein structures to predict whether a given protein sequence can be expressed successfully in E. coli., achieving state-of-the-art performance.
At the ACS conference, Collin unveiled findings from molecular dynamics simulations, examining the interactions between fluorescent probes and linoleic acid hydroperoxide (LinAOOH) within a lipid bilayer model. This research offers vital insights into the positions and orientations of multiple fluorescent probes and LinAOOH shedding light on their potential oxidation mechanisms.
At ACS Fall 23, Hojin presented a scheme for obtaining self-adaptive solubility databases using semi-supervised distillation and graph neural networks to predict solubility. With the scheme, solubility prediction can be successfully applied to estimating reaction rates of organic reactions and predicting partition coefficients of lignin derivatives and drug-like molecules.