Data-driven discovery of molecular photoswitches with multioutput Gaussian processes

Photoswitchable molecules display two or more isomeric forms that may be accessed using light. Separating the electronic absorption bands of these isomers is key to selectively addressing a specific isomer and achieving high photostationary states whilst overall red-shifting the absorption bands serves to limit material damage due to UV-exposure and increases penetration depth in photopharmacological applications. Engineering these properties into a system through synthetic design however, remains a challenge. Here, we present a data-driven discovery pipeline for molecular photoswitches underpinned by dataset curation and multitask learning with Gaussian processes. In the prediction of electronic transition wavelengths, we demonstrate that a multioutput Gaussian process (MOGP) trained using labels from four photoswitch transition wavelengths yields the strongest predictive performance relative to single-task models as well as operationally outperforming time-dependent density functional theory (TD-DFT) in terms of the wall-clock time for prediction. We validate our proposed approach experimentally by screening a library of commercially available photoswitchable molecules. Through this screen, we identified several motifs that displayed separated electronic absorption bands of their isomers, exhibited red-shifted absorptions, and are suited for information transfer and photopharmacological applications. Our curated dataset, code, as well as all models are made available at https://github.com/Ryan-Rhys/The-Photoswitch-Dataset.

We present a data-driven discovery pipeline for molecular photoswitches through multitask learning with Gaussian processes. Through subsequent screening, we identify several motifs with separated and red-shifted electronic absorption bands.     

FragNet,a Contrastive Learning-Based Transformer Model for Clustering,Interpreting, Visualizing,and Navigating Chemical Space

The question of molecular similarity is core in cheminformatics and is usually assessed via a pairwise comparison based on vectors of properties or molecular fingerprints. We recently exploited variational autoencoders to embed 6M molecules in a chemical space, such that their (Euclidean) distance within the latent space so formed could be assessed within the framework of the entire molecular set. However, the standard objective function used did not seek to manipulate the latent space so as to cluster the molecules based on any perceived similarity. Using a set of some 160,000 molecules of biological relevance, we here bring together three modern elements of deep learning to create a novel and disentangled latent space, viz transformers, contrastive learning, and an embedded autoencoder. The effective dimensionality of the latent space was varied such that clear separation of individual types of molecules could be observed within individual dimensions of the latent space. The capacity of the network was such that many dimensions were not populated at all. As before, we assessed the utility of the representation by comparing clozapine with its near neighbors, and we also did the same for various antibiotics related to flucloxacillin. Transformers, especially when as here coupled with contrastive learning, effectively provide one-shot learning and lead to a successful and disentangled representation of molecular latent spaces that at once uses the entire training set in their construction while allowing “similar” molecules to cluster together in an effective and interpretable way.     

Oxygen concentration and modeling thermal decomposition of a high‐performance material: A case study of polyimide (Cirlex)

Kinetic decomposition models for the thermal decomposition of a high‐performance polymeric material (Polyimide, PI) were determined from specific techniques. Experimental data from thermogravimetric analysis (TGA) and previously elucidated decomposition mechanism were combined with numerical simulating tool to establish a comprehensive kinetic model for the decomposition of PI under three atmospheres: nitrogen, 2% oxygen, and synthetic air. Multistaged kinetic models with subsequent and competitive reactions were established by taking into consideration the different types of reactions that may be occurring during the thermal decomposition of the material (chain scission, thermo‐oxidation, char formation). The decomposition products and decomposition mechanism of PI which was established in our previous report allowed for the elucidation of the kinetic decomposition models. A three‐staged kinetic thermal decomposition pathway was a good fit to model the thermal decomposition of PI under nitrogen. The kinetic model involved an autocatalytic type of reaction followed by successive nth order reactions. Such types of models were set up for the evaluation of the kinetics of the thermal decomposition of PI under 2% oxygen and in air, leading to models with satisfactory fidelity.     

Accelerated Repurposing and Drug Development of Pulmonary Hypertension Therapies for COVID-19 Treatment Using an AI-Integrated Biosimulation Platform

The COVID-19 pandemic has reached over 100 million worldwide. Due to the multi-targeted nature of the virus, it is clear that drugs providing anti-COVID-19 effects need to be developed at an accelerated rate, and a combinatorial approach may stand to be more successful than a single drug therapy. Among several targets and pathways that are under investigation, the renin-angiotensin system (RAS) and specifically angiotensin-converting enzyme (ACE), and Ca²⁺-mediated SARS-CoV-2 cellular entry and replication are noteworthy. A combination of ACE inhibitors and calcium channel blockers (CCBs), a critical line of therapy for pulmonary hypertension, has shown therapeutic relevance in COVID-19 when investigated independently. To that end, we conducted in silico modeling using BIOiSIM, an AI-integrated mechanistic modeling platform by utilizing known preclinical in vitro and in vivo datasets to accurately simulate systemic therapy disposition and site-of-action penetration of the CCBs and ACEi compounds to tissues implicated in COVID-19 pathogenesis.     

Biochemical and Biophysical Characterisation of the Hepatitis E Virus Guanine-7-Methyltransferase

Hepatitis E virus (HEV) is an understudied pathogen that causes infection through fecal contaminated drinking water and is prominently found in South Asian countries. The virus affects ~20 million people annually, leading to ~60,000 infections per year. The positive-stranded RNA genome of the HEV genotype 1 has four conserved open reading frames (ORFs), of which ORF1 encodes a polyprotein of 180 kDa in size, which is processed into four non-structural enzymes: methyltransferase (MTase), papain-like cysteine protease, RNA-dependent RNA polymerase, and RNA helicase. MTase is known to methylate guanosine triphosphate at the 5′-end of viral RNA, thereby preventing its degradation by host nucleases. In the present study, we cloned, expressed, and purified MTase spanning 33–353 amino acids of HEV genotype 1. The activity of the purified enzyme and the conformational changes were established through biochemical and biophysical studies. The binding affinity of MTase with magnesium ions (Mg²⁺) was studied by isothermal calorimetry (ITC), microscale thermophoresis (MST), far-UV CD analysis and, fluorescence quenching. In summary, a short stretch of nucleotides has been cloned, coding for the HEV MTase of 37 kDa, which binds Mg²⁺ and modulate its activity. The chelation of magnesium reversed the changes, confirming its role in enzyme activity.     

Parallel Correlation Studies of Fluctuations and the Impact of Perturbations in the Magnetic Configuration

The plasma turbulence in the boundary of fusion relevant experiments is known to have a quasi two‐dimensional nature: the scale lengths perpendicular to the magnetic field are in the order of mm to cm, but parallel to the magnetic field, the correlation lengths are in the order of several meters. Recent parallel correlation studies with Langmuir probes at the JET tokamak over very long connection lengths (23 m and 66 m probe tip separation along the magnetic field) showed a correlation of less than 50%, in contrast to the finding of 80–90% correlation in other devices at measurements with smaller probe tip separations. However, it was not clear if this is a genuine physical property of the electrostatic turbulence in the scrape‐off layer or whether perturbations in the magnetic configuration had caused an additional decorrelation by a time‐dependent misalignment of the two probe tips along the connecting field line. In this contribution we analyze the effect of such perturbations in the magnetic configuration on the parallel correlation measurements in a simple model and compare the results with those of the measurements at JET.     

Simplified model and analysis of a pair of coupled thermo-optical MEMS oscillators

Nonlinear Dynamics - Motivated by the dynamics of microscale oscillators with thermo-optical feedback, a simplified third-order model capturing the key features of these oscillators is developed,...