arXiv subjects

Faranak Hatami

Publications and source records attributed to Faranak Hatami.

4 recordsLinked to original sources

arXiv2026

Asymmetric Focal Loss Improves Graph Neural Network Prediction of Drug-Drug Interactions

Background: Graph neural networks improve computational prediction of polypharmacy side effects, but standard binary cross-entropy training allocates equal capacity to well-classified and difficult examples, potentially missing clinically significant interactions. We evaluated whether an asymmetric focal objective could improve multi-relational drug-drug interaction (DDI) prediction by emphasizing difficult positive interactions. Methods: ClinicalFocal loss was integrated into a relation-aware graph convolutional network using molecular fingerprints, physicochemical descriptors, and learned embeddings. The model was evaluated on TWOSIDES using five-fold cross-validation with identical experimental conditions (architecture, features, data partitions, hyperparameters, and random seeds) for ClinicalFocal loss and binary cross-entropy baseline. Results: ClinicalFocal loss increased accuracy from 0.699 to 0.892 (+19.3 percentage points) and F1 score from 0.700 to 0.894 (+19.4 percentage points). AUROC increased from 0.766 to 0.914, and AUCPR increased from 0.714 to 0.860. The false-negative rate decreased from 29.8% to 9.1%, while specificity increased from 69.6% to 87.5%. Overall classification error decreased from 30.1% to 10.8%, corresponding to a 64.1% relative reduction. Improvements were consistent across all five folds. Conclusions: Asymmetric focal optimization improved classification and ranking performance while achieving 90.9% recall for observed interaction triples, without modifying the underlying architecture. Loss-function design is a direct, tunable lever for improving graph-based DDI prediction.

cs.LG↗

arXiv2026

Molecular Dynamics Force Field Genetic Optimization for Tri-n-butyl Phosphate Liquid

An iterative optimization algorithm with MD simulations in the loop is developed and applied to optimize Lennard-Jones (LJ) parameters specific for liquid tri-n-butyl phosphate (TBP). The optimization loop uses non-dominated sorting genetic algorithms to obtain LJ parameters that reproduce key properties such as mass density, electric dipole moment, heat of vaporization, self-diffusion coefficient (SDC), and shear viscosity. Errors relative to experimentally measured properties lead to a multi-objective function optimization problem stated in terms of a Pareto-optimal set. A systematic application of the optimization algorithm to cases involving single- and multi-objective functions was carried out in this work, establishing a framework for atomistic TBP property predictions. We demonstrate the use of a neural network property model to amortize the high cost of MD simulations in the optimization loop and to allow for large populations and more generations to be used in the genetic algorithms. In our previous study of finding the best force field for TBP property predictions as judged by the aforementioned thermophysical properties, we found the Polarized AMBER-MNDO force field to be the best overall showing a \num{74}\% relative deviation from experimental values. However, in this study, we show optimized values of the LJ parameters that improve the overall deviation from experimental data to \num{23}\% when using the NN NSGA-III algorithm. Despite this large improvement, the accurate prediction of the transport properties, SDC and shear viscosity, remains difficult since improvements in one of them worsen the other, and vice versa.

physics.chem-ph↗

arXiv2024

Energy Spectrum of Primary Knock-on Atoms and Atomic Displacement Calculations in Metallic Alloys Under Neutron Irradiation

Materials subjected to neutron irradiation experience damage due to displacement cascades triggered by nuclear reactions. This paper presents a practical method to calculate primary atomic recoil events (PKAs), which lead to cascade damage, based on energy and recoiling species. We developed a custom code to identify PKAs and extract their properties using MCNPX and SRIM. This code determines the specifications of recoil atoms from the data provided by the PTRAC card in MCNPX. Consequently, the energy spectrum of PKAs generated through various reaction channels, including elastic/inelastic scattering and transmutations such as (n, {\alpha}), (n, p), and (n, {\gamma}), is calculated. This PKA spectrum is then input into SRIM, which calculates the total number of atomic displacements using the binary collision approximation (BCA) and provides crucial information about the spatial distribution of defects within the irradiated material. Our results indicate that elastic scattering is the predominant reaction, producing most PKAs with energies in the range of several keV. In contrast, inelastic scattering becomes the dominant interaction for generating high-energy PKAs (~EPKA>1 MeV). Additionally, we observed that the number of Frenkel pairs versus ion energy curves for light particle ion implantation (such as H and He) is significantly smaller than for heavier ions.

physics.comp-ph↗

arXiv2023

Comparison of Different Machine Learning Approaches to Predict Viscosity of Tri-n-Butyl Phosphate Mixtures Using Experimental Data

Tri-n-butyl phosphate (TBP) is a solvent that is commonly used in a variety of industries, including the nuclear and chemical industries, for its ability to dissolve and purify various inorganic acids and metals. It is often used in hydrometallurgical processes to separate and purify these substances. Machine learning models offer a promising alternative to traditional methods for predicting the viscosity of TBP mixtures. By training machine learning models on a dataset of viscosity measurements, it is possible to accurately predict the viscosity of TBP mixtures at different compositions, densities, and temperatures, which can save time and resources and reduce the risk of exposure to toxic solvents. This paper aimed at proposing Machine Learning (ML) techniques to automatically predict the viscosity of TBP mixtures using experimental data. For comparison peruses, we trained five different ML algorithms including Support Vector Regressor (SVR), Random Forest (RF), Logistic Regression (LR), Gradient Boosted Decision Trees (XGBoost), and Neural Network (NN). We collected a total of 511 measurements for TBP mixtures with temperature-based density, at different compositions, containing hexane, dodecane, cyclohexane, n-heptane, toluene, and ethylbenzene measured at temperatures of T= (288.15, 293.15, 298.15, 303.15, 308.15, 313.15, 318.15, 323.15, and 328.15) K. The results revealed that the NN model with 25 and 50 neurons in the hidden layers could achieve the best viscosity predictions for a system of TBP mixtures. The NN model outperformed other regular ML models in terms of Mean Square Error (MSE) of 0.157 % and adjusted R2 of 99.72 % on the test data set. This paper demonstrated that the NN model can be an appropriate option to accurately predict the viscosity of TBP + Ethylbenzene with a margin of deviation as low as 0.049 %.

physics.comp-ph↗