arXiv subjects

Thomas Joseph

Publications and source records attributed to Thomas Joseph.

3 recordsLinked to original sources

arXiv2025

MIDOG 2025 Track 2: A Deep Learning Model for Classification of Atypical and Normal Mitotic Figures under Class and Hardness Imbalances

Motivation: Accurate classification of mitotic figures into normal and atypical types is crucial for tumor prognostication in digital pathology. However, developing robust deep learning models for this task is challenging due to the subtle morphological differences, as well as significant class and hardness imbalances in real-world histopathology datasets. Methods: We propose a novel deep learning approach based on a ResNet backbone with specialized classification heads. Our architecture uniquely models both the mitotic figure phenotype and the instance difficulty simultaneously. This method is specifically designed to handle the challenges of diverse tissue types, scanner variability, and imbalanced data. We employed focal loss to effectively mitigate the pronounced class imbalance, and a comprehensive data augmentation pipeline was implemented to enhance the model's robustness and generalizability. Results: Our approach demonstrated strong and consistent performance. In a 5-fold cross-validation on the MIDOG 2025 Track 2 dataset, it achieved a mean balanced accuracy of 0.8744 +/- 0.0093 and an ROC AUC of 0.9505 +/- 0.029. The model showed robust generalization across preliminary leaderboard evaluations, achieving an overall balanced accuracy of 0.8736 +/- 0.0204. Conclusion: The proposed method offers a reliable and generalizable solution for the classification of atypical and normal mitotic figures. By addressing the inherent challenges of real world data, our approach has the potential to support precise prognostic assessments in clinical practice and improve consistency in pathological diagnosis.

eess.IV↗

arXiv2018

A streamlined, general approach for computing ligand binding free energies and its application to GPCR-bound cholesterol

The theory of receptor-ligand binding equilibria has long been well-established in biochemistry, and was primarily constructed to describe dilute aqueous solutions. Accordingly, few computational approaches have been developed for making quantitative predictions of binding probabilities in environments other than dilute isotropic solution. Existing techniques, ranging from simple automated docking procedures to sophisticated thermodynamics-based methods, have been developed with soluble proteins in mind. Biologically and pharmacologically relevant protein-ligand interactions often occur in complex environments, including lamellar phases like membranes and crowded, non-dilute solutions. Here we revisit the theoretical bases of ligand binding equilibria, avoiding overly specific assumptions that are nearly always made when describing receptor-ligand binding. Building on this formalism, we extend the asymptotically exact Alchemical Free Energy Perturbation technique to quantifying occupancies of sites on proteins in a complex bulk, including phase-separated, anisotropic, or non-dilute solutions, using a thermodynamically consistent and easily generalized approach that resolves several ambiguities of current frameworks. To incorporate the complex bulk without overcomplicating the overall thermodynamic cycle, we simplify the common approach for ligand restraints by using a single distance-from-bound-configuration (DBC) ligand restraint during AFEP decoupling from protein. DBC restraints should be generalizable to binding modes of most small molecules, even those with strong orientational dependence. We apply this approach to compute the likelihood that membrane cholesterol binds to known crystallographic sites on 3 GPCRs at a range of concentrations. Non-ideality of cholesterol in a binary cholesterol:POPC bilayer is characterized and consistently incorporated into the interpretation.

q-bio.QM↗

arXiv2014

Smart city analysis using spatial data and predicting the sustainability

Smart city [1] planning is crucial as it should balance among resources and the needs of the city .It allows to achieve good eco-friendly industries, there by supporting both the nature and the stake holders. Setting up an industry is a difficult problem, because it should optimize the resources and allocating it in an effective manner. Weighted sum approach [2] uses the spatial data for finding appropriate places to set up the industry based on the weight assigned to each constraint. The user can predict the possible places in the search space, where the industry can be set with low time complexity using spatial data. Diversity being introduced by using multipoint crossover and mutation operations. It will help to bring exploration in the search space, thereby bring the diversity factor into the solution space. The prediction approach will help to avoid the human exploitation on nature for resources. This in turn helps the investors to maximize the Return on Investment (ROI).

cs.CY↗