Exploring computational pathology · multi-omics · multimodal AI to understand cancer and biological systems.
This site is a lightweight reading notebook: concise, opinionated notes on papers I read, with an emphasis on how ideas can be turned into robust analysis pipelines rather than only benchmark numbers.
Skimmed a paper on aligning histology with bulk RNA-seq using a joint latent space. Noted their loss design and how they handle confounders. Wrote down a potential idea for integrating perturbation data into a similar framework.
Read a large-scale CRISPR perturbation study. Recorded how they model gene-gene interactions and how that might be combined with spatial information.
Compared several nuclei instance segmentation architectures. Reflected on continual learning and how to design an interactive annotation loop.
Compared several nuclei instance segmentation architectures. Reflected on continual learning and how to design an interactive annotation loop.
Compared several nuclei instance segmentation architectures. Reflected on continual learning and how to design an interactive annotation loop.
Compared several nuclei instance segmentation architectures. Reflected on continual learning and how to design an interactive annotation loop.
A brief summary of the problem setting, integration strategy, and how this work connects to computational pathology and cancer subtyping...
Notes on a study using large-scale perturbation screens to infer gene regulatory programs and cellular responses, and how these signals can be aligned with other omics layers...
A reflection on different architectures for nuclei instance segmentation, how they handle overlapping/touching nuclei, and some ideas about continual learning and prompts...
Notes on models and biomarkers used for cancer diagnosis from histopathology, with comments on robustness, domain shift, and interpretability...
Summary of survival models, risk scores, and how omics + image features are combined to predict outcomes...
A note on a foundation model for histopathology images: pretraining strategy, downstream tasks, scaling behavior, and open questions for multi-omics and spatial data...
Thoughts on models that jointly learn from images, text, and omics, and how such representations can support reasoning and downstream tasks...