Biology subjects

Bhargava, N.

Publications and source records attributed to Bhargava, N..

3 recordsLinked to original sources

bioRxiv2026

Decoupled Representation Learning Improves Generalization in CRISPR Off-Target Prediction

BackgroundComputational prediction of CRISPR-Cas9 off-target activity is essential for safe guide-RNA design, yet models trained on large proxy datasets often fail to generalize to experimentally validated sites. MethodsWe present a modular two-stage deep learning framework that separates sequence representation learning from off-target classification. In Stage 1, guide RNA sequences are encoded using frozen, pretrained DNABERT embeddings learned from large genomic corpora. In Stage 2, these embeddings are integrated with mismatch-level and pairwise sequence features within a hybrid CNN-Transformer classifier trained exclusively on a high-throughput proxy dataset. ResultsOn the external TrueOT benchmark, a curated collection of low-throughput, experimentally confirmed off-target sites, the full model achieved a mean ROC-AUC of 0.70 {+/-} 0.03 and a PR-AUC of 0.30 {+/-} 0.03, markedly surpassing the proxy-only baseline (ROC-AUC=0.64, PR-AUC=0.22). Ablation studies confirmed that the performance gain arises from the pretrained sequence representations rather than architectural complexity. ConclusionsDecoupling representation learning from downstream classification and leveraging frozen transformer-based embeddings substantially improves generalization to biologically relevant off-target predictions. The proposed framework provides a reproducible baseline for the assessment of CRISPR-Cas9 risk and underscores the importance of transfer learning in the integration of proxy test data and experimental results in the real-world.

bioinformatics↗

bioRxiv2023

Machine learning models for prediction of xenobiotic chemicals with high propensity to transfer into human milk

Breast milk serves as a vital source of essential nutrients for infants. However, human milk contamination via transfer of environmental chemicals from maternal exposome is a significant concern for infant health. Machine learning based predictive toxicology models can be valuable in predicting chemicals with high propensity to transfer into human milk. To this end, we build such classification- and regression-based models by employing multiple machine learning algorithms and leveraging the largest curated dataset to date of 375 chemicals with known Milk to Plasma concentration (M/P) ratios. Our Support Vector Machine (SVM) based classifier outperforms other models in terms of different performance metrics, when evaluated on both (internal) test data and external test dataset. Specifically, the SVM based classifier on (internal) test data achieved a classification accuracy of 77.33%, specificity of 84%, sensitivity of 64%, and F-score of 65.31%. When evaluated on an external test dataset, our SVM based classifier is found to be generalizable with sensitivity of 77.78%. While we were able to build highly predictive classification models, our best regression models for predicting the M/P ratio of chemicals could achieve only moderate R2 values on the (internal) test data. As noted in earlier literature, our study also highlights the challenges in developing accurate regression models for predicting the M/P ratio of xenobiotic chemicals. We have made our complete workflow, train and test datasets, and computer codes for the classification and regression models publicly available via a dedicated GitHub repository. Overall, this study attests the immense potential of predictive computational toxicology models in characterizing the myriad chemicals in the human exposome.

pharmacology and toxicology↗

bioRxiv2021

Development of an efficient single-cell cloning and expansion strategy for genome edited induced pluripotent stem cells

Disease-specific human induced pluripotent stem cells (hiPSCs) can be generated directly from individuals with known disease characteristics or alternatively be modified using genome editing approaches to introduce disease causing genetic mutations to study the biological response of those mutations. The genome editing procedure in hiPSCs is still inefficient, particularly when it comes to homology directed repair (HDR) of genetic mutations or targeted transgene insertion in the genome and single cell cloning of edited cells. In addition, genome editing processes also involve additional cellular stresses such as trouble with cell viability and genetic stability of hiPSCs. Therefore, efficient workflows are desired to increase genome editing application to hiPSC disease models and therapeutic applications. Apart from genome editing efficiency, hiPSC survival following single-cell cloning has proved to be challenging and has thus restricted the capability to easily isolate homogeneous clones from edited hiPSCs. To this end, we demonstrate an efficient workflow for feeder-free single cell clone generation and expansion in both CRISPR-mediated knock-out (KO) and knock-in (KI) hiPSC lines. Using StemFlex medium and CloneR supplement in conjunction with Matrigel cell culture matrix, we show that cell viability and expansion during single-cell cloning in edited and unedited cells is significantly enhanced. Our reliable single-cell cloning and expansion workflow did not affect the biology of the hiPSCs as the cells retained their growth and morphology, expression of various pluripotency markers and normal karyotype. This simplified and efficient workflow will allow for a new level of sophistication in generating hiPSC-based disease models to promote rapid advancement in basic research and also the development of novel cellular therapeutics.

cell biology↗