arXiv subjects

Lydia Ng

Publications and source records attributed to Lydia Ng.

3 recordsLinked to original sources

arXiv2025

STRUCTSENSE: A Task-Agnostic Agentic Framework for Structured Information Extraction with Human-In-The-Loop Evaluation and Benchmarking

Extracting structured information from scientific literature is critical for accelerating discovery, yet Large Language Models (LLMs) often struggle in specialized domains that require expert knowledge and generalize poorly across tasks. We introduce \textsc{StructSense}, a modular, task-agnostic, open-source framework that integrates ontology-guided symbolic knowledge, agentic self-evaluative refinement, and human-in-the-loop validation for robust domain-aware extraction. We evaluate \textsc{StructSense} on three tasks of increasing semantic complexity: schema-based extraction of assessment instruments (91--100\% accuracy), metadata and resource extraction from scientific papers (86--93\% overall), and named entity recognition (NER) from neuroscience literature (58--75\% label accuracy across 8,882 entities). On two biomedical NER benchmarks (NCBI Disease and S800 Species), the system achieves $\geq$90\% relaxed recall and 62.5--85.8\% strict recall while extracting 1,000--3,600 additional entities beyond gold annotations. The local concept mapping service achieves Hits@1 of 62--82\% under strict matching and 68--86\% under semantic matching. These results across three domains demonstrate that \textsc{StructSense} generalizes across tasks while maintaining source grounding and provenance transparency.

cs.CL↗

arXiv2014

Cell-type-specific transcriptomes and the Allen Atlas (II): discussion of the linear model of brain-wide densities of cell types

The voxelized Allen Atlas of the adult mouse brain (at a resolution of 200 microns) has been used in [arXiv:1303.0013] to estimate the region-specificity of 64 cell types whose transcriptional profile in the mouse brain has been measured in microarray experiments. In particular, the model yields estimates for the brain-wide density of each of these cell types. We conduct numerical experiments to estimate the errors in the estimated density profiles. First of all, we check that a simulated thalamic profile based on 200 well-chosen genes can transfer signal from cerebellar Purkinje cells to the thalamus. This inspires us to sub-sample the atlas of genes by repeatedly drawing random sets of 200 genes and refitting the model. This results in a random distribution of density profiles, that can be compared to the predictions of the model. This results in a ranking of cell types by the overlap between the original and sub-sampled density profiles. Cell types with high rank include medium spiny neurons, several samples of cortical pyramidal neurons, hippocampal pyramidal neurons, granule cells and cholinergic neurons from the brain stem. In some cases with lower rank, the average sub-sample can have better contrast properties than the original model (this is the case for amygdalar neurons and dopaminergic neurons from the ventral midbrain). Finally, we add some noise to the cell-type-specific transcriptomes by mixing them using a scalar parameter weighing a random matrix. After refitting the model, we observe than a mixing parameter of $5\%$ leads to modifications of density profiles that span the same interval as the ones resulting from sub-sampling.

q-bio.NC↗

arXiv2012

A cell-type based model explaining co-expression patterns of genes in the brain

Much of the genome is expressed in the vertebrate brain, with individual genes exhibiting different spatially-varying patterns of expression. These variations are not independent, with pairs of genes exhibiting complex patterns of co-expression, such that two genes may be similarly expressed in one region, but differentially expressed in other regions. These correlations have been previously studied quantitatively, particularly for the gene expression atlas of the mouse brain, but the biological meaning of the co-expression patterns remains obscure. We propose a simple model of the co-expression patterns in terms of spatial distributions of underlying cell types. We establish the plausibility of the model in terms of a test set of cell types for which both the gene expression profiles and the spatial distributions are known.

q-bio.QM↗