Biology subjects

Kay, J.

Publications and source records attributed to Kay, J..

4 recordsLinked to original sources

bioRxiv2026

Scrub Data: A Framework for Reproducible Data Curation with AI Coding Agents

Data curation-the process of collecting, cleaning, and joining raw datasets for unified analysis-is a time-consuming yet crucial part of any data science project. With the emergence of powerful agentic coding tools, it is tempting to "vibe curate"-i.e., to prompt AI agents to perform data curation operations and blindly trust their outputs in order to finish work more quickly. However, this can exacerbate two key challenges that already plague manual data curation workflows: reproducibility-the ability to trace the exact set of modifications applied to raw datasets-and verifiability-the ability to audit changes and confirm that data is processed properly. To address these issues, we introduce Scrub Data, a framework that enables the use of AI agents in data curation workflows while transparently maintaining both reproducibility and verifiability. The framework involves an interactive data curation loop between a user and an AI agent, backed by a data provenance graph that tracks and versions all dataset updates. The data graph requires that all updates are formatted as self-contained executable steps, allowing any version of a dataset to be reconstructed from raw data by playing forward the transformations stored in the graph. This workflow takes place within a lightweight web application that is designed to be modified by users and agents in real-time to create custom visualization tools for identifying data curation needs and verifying outcomes. We demonstrate the usefulness of the framework with a real data curation use case from animal movement ecology. Starting with 89M raw GPS coordinates, we use the framework to curate a benchmark dataset of 2.5M coordinates to be used for machine learning or statistical analysis. The framework is available for use and extension at https://github.com/justinkay/scrubdata.

ecology↗

bioRxiv2026

Novel Methylation Markers in a Prostate Cancer Cohort are Associated with Disease Development and Relapse

Prostate cancer remains one of the most common cancers among men globally. While significant strides have been made in diagnosis and treatment, understanding the complex genetic and epigenetic underpinnings of the disease remains crucial for guiding intervention and developing more personalized and effective therapies. The importance of DNA methylation in prostate cancer has been known for some time, but important facets of the modulation of the epigenome during carcinogenesis remain obscure, partly because the bulk of cancer methylation data have been produced using microarray technologies. Here we utilise the TruSeq methyl capture method (EPICseq) to profile the, previously defined, UK Prostate ICGC cohort of well-annotated primary prostate cancers. To this we add methylation sequencing of benign tissue from the same men. These data allow us to identify differentially methylated regions distinguishing cancerous and non-cancerous prostate tissue, while identifying numerous genes whose methylation profiles can perform that task as well as distinguishing between classes of prostate cancer. We describe a describe a methylation-based control mechanism for prostate-cancer-associated SNPs, and show that this seems a likely mechanism of action for a SNP near the MMP7 gene. We describe three novel molecular signatures that arise from different aspects of the biology of prostate cancer revealed by sequencing. Each is shown to be an independent classifier of cancers into groups with different expected times to relapse. These consist of patterns in driver gene methylation, strand-specific methylation, and signal arising in mitochondrial reads. We show that these signatures, combined with existing molecular tools, provide a powerful predictor of time to recurrence. By substantially enhancing understanding of prostate cancer risk, detection, and prognosis, we pave the way for the development of clinical practices that will benefit patients and improve outcomes.

cancer biology↗

bioRxiv2025

Control of two-pathway signal integration in a model neocortical pyramidal cell

We demonstrate how neuromodulation and spatially targetted inhibition can alter the integration of the two streams of excitatory input in a thick-tufted layer 5 pyramidal cell, using a computational reduced-compartmental cell models. Choosing suitable ranges of brief current stimulus amplitudes, applied basally to either the soma or basal dendrites (basal stimulation) and to the apical tuft (apical stimulation) results in burst firing due to either stimulus alone, if strong enough, or by a combination of the stimuli at lower amplitudes. Applying tonic inhibition to the apical tuft removes the ability of apical input alone to generate a burst over the chosen amplitude range. A similar effect is achieved by reducing the tuft calcium channel conductance as an outcome of neuromodulation. Similarly, tonic inhibition to the basal dendrites removes the ability of basal stimulation alone to generate a burst, without blocking bursts resulting from apical calcium spikes. HCN channels in the apical dendrites may amplify or reduce bursting probability, depending on other active and passive properties of dendrites. So neuromodulation that decreases the conductance of these channels may act to reduce or increase bursting probability across the across the ranges of basal and apical inputs, depending on cell properties. These effects mimic those found previously by simply limiting the range of stimulus amplitudes (Graham et al., 2025) but now show that such changes in two-stream signal integration can happen through network inhibition and neuromodulation with no change in the excitatory driving stimulus strengths. These changes in signal integration also lead to changes in information transmitted by the cells bursting probability about the two input streams, as shown in Graham et al. (2025). Changes in cell morphology are also investigated by reducing the apical trunk length and are revealed to alter this two-stream signal integration through differential effects on passive and active interaction between the soma and apical tuft.

neuroscience↗

bioRxiv2019

A Bayesian decomposition of BAC firing as a mechanism for apical amplification in neocortical pyramidal neurons

Pyramidal cells in layer 5 of the neocortex have two distinct integration sites. These cells integrate inputs to basal dendrites in the soma while integrating inputs to the tuft in a site at the top of the apical trunk. The two sites communicate by action potentials that backpropagate to the apical site and by backpropagation-activated calcium spikes (BAC firing) that travel from the apical to the somatic site. Six key messages arise from the probabilistic information-theoretic analyses of BAC firing presented here. First, we suggest that pyramidal neurons with BAC firing could convert the odds in favour of the presence of a feature given the basal data into the odds in favour of the presence of a feature given the basal data and the apical input, by a simple Bayesian calculation. Second, the strength of the cells response to basal input can be amplified when relevant to the current context, as specified by the apical input, without corrupting the message that it sends. Third, these analyses show rigorously how this apical amplification depends upon communication between the sites. Fourth, we use data on action potentials from a very detailed multi-compartmental biophysical model to study our general model in a more realistic setting, and demonstrate that it describes the data well. Fifth, this form of BAC firing meets criteria for distinguishing modulatory from driving interactions that have been specified using recent definitions of multivariate mutual information. Sixth, our general decomposition can be extended to cases where, instead of being purely driving or purely amplifying, apical and basal inputs can be partly driving and partly amplifying to various extents. These conclusions imply that an advance beyond the assumption of a single site of integration within pyramidal cells is needed, and suggest that the evolutionary success of neocortex may depend upon the cellular mechanisms of context-sensitive selective amplification hypothesized here.\n\nAuthor summaryThe cerebral cortex has a key role in conscious perception, thought, and action, and is predominantly composed of a particular kind of neuron: the pyramidal cells. The distinct shape of the pyramidal neuron with a long dendritic shaft separating two regions of profuse dendrites allows them to integrate inputs to the two regions separately and combine the results non-linearly to produce output. Here we show how inputs to this more distant site strengthen the cells output when it is relevant to the current task and environment. By showing that such neurons have capabilities that transcend those of neurons with the single site of integration assumed by many neuroscientists, this splitting of the neuronal atom offers a radically new viewpoint from which to understand the evolution of the cortex and some of its many pathologies. This also suggests that approaches to artificial intelligence using neural networks might come closer to something analogous to real intelligence, if, instead of basing them on processing elements with a single site of integration, they were based on elements with two sites, as in cortex.

neuroscience↗