Watson, O. | bioRxiv Science

bioRxiv2025

Transfer Learning Assessment of Data-driven Crystallisation Processes via Constrained Neural Ordinary Differential Equations

Modelling complex crystallisation processes remains challenging due to limited experimental datasets, high measurement noise, and the need for generalisability across varying operating conditions. Neural Ordinary Differential Equations (NODEs) and transfer learning (TL) offer promising tools to overcome these limitations by providing data-efficient, flexible, and transferable modelling frameworks. This work investigates the use of NODEs to model protein crystallisation dynamics under data-scarce conditions. A NODE trained on a data-rich source system successfully captures solute consumption and particle size dynamics, but when applied to data-sparse target systems, scratch-trained NODEs exhibit limited generalisation and unphysical behaviours. To address this, several TL strategies are evaluated, including layer freezing, parameter deviation penalisation, and system-embedding within the neural architecture. Results show that layer freezing and deviation penalty consistently improve knowledge transfer, while system-embedding offers robustness in noisy or undersampled datasets. In addition, physics-informed NODEs, constrained to enforce monotonic concentration decay and crystal growth, demonstrate greater stability under high noise and sparse measurement regimes, ensuring physically consistent predictions. Overall, the combination of constrained NODEs with appropriate TL strategies provides a robust framework for accurate, transferable modelling of crystallisation systems in low-data regimes.

bioengineering↗

bioRxiv2022

Examining the early distribution of the artemisinin-resistant Plasmodium falciparum kelch13 R561H mutation in Rwanda

Artemisinin resistance mutations in Plasmodium falciparum kelch13 (Pfk13) have begun to emerge in Africa. Pfk13-R561H was the first reported African mutation found in Rwanda in 2014, but limited sampling left questions about its early distribution and origin. We detected 476 parasitemias among 1873 residual blood spots from a 2014-15 Rwanda Demographic Health Survey. We sequenced 351 samples revealing 341/351 were wild type (97.03% weighted) and 4 samples (1.34% weighted) harbored R561H which were significantly spatially clustered. Our study better defines the early distribution of R561H in Rwanda and suggests that the origin may have involved higher-transmission regions.

genomics↗

bioRxiv2019

Riluzole does not ameliorate disease caused by cytoplasmic TDP-43 in a mouse model of amyotrophic lateral sclerosis

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease commonly treated with riluzole, a small molecule that may act via modulation of glutamatergic neurotransmission. However, riluzole only modestly extends lifespan for people living with ALS and its precise mechanisms of action remain unclear. Most ALS cases are characterised by accumulation of cytoplasmic TAR DNA binding protein of 43 kDa (TDP-43), and understanding the effects of riluzole in models that closely recapitulate TDP-43 pathology may provide insights for development of improved therapeutics. We therefore investigated the effects of riluzole in transgenic mice that inducibly express nuclear localisation sequence (NLS)-deficient human TDP-43 in neurons (NEFH-tTA/tetO-hTDP-43{Delta}NLS, rNLS, mice). Riluzole treatment from the first day of hTDP-43{Delta}NLS expression did not alter disease onset, weight loss or performance on multiple motor behavioural tasks. Riluzole treatment also did not alter TDP-43 protein levels, solubility or phosphorylation. Although we identified a significant decrease in GluA2 and GluA3 proteins in the cortex of rNLS mice, riluzole did not ameliorate this disease-associated molecular phenotype. Likewise, riluzole did not alter the disease-associated atrophy of hindlimb muscle in rNLS mice. Finally, riluzole treatment beginning after disease onset in rNLS mice similarly had no effect on progression of late-stage disease or animal survival. Together, we demonstrate specific glutamatergic receptor alterations and muscle fibre-type changes reminiscent of ALS in rNLS mice, but riluzole had no effect on these or any other disease phenotypes. Future targeting of pathways directly related to accumulation of TDP-43 pathology may be needed to develop better treatments for ALS. Significance StatementAccumulation of cytoplasmic TDP-43 protein is the hallmark pathology of ALS. Riluzole is the most widely used drug for ALS treatment, but provides only a short extension of lifespan. We demonstrate here in the rNLS mouse model, which mimics TDP-43 pathology, that riluzole does not ameliorate progressive alterations in motor strength and coordination, muscle atrophy, glutamate receptor levels, or TDP-43 protein levels and solubility, and does not prolong animal survival. Riluzole similarly did not affect decreased levels of glutamate receptor subunits GluA2/GluA3 in rNLS mice. The inability of riluzole to rescue pathological or phenotypic changes in this TDP-43 model provides further impetus for the discovery of improved therapies targeting the key drivers of ALS pathogenesis.

neuroscience↗