CATO

Research Tools

CATO runs Python and R on your project's computer and saves the notebooks, code, and results.

Ask in scientific terms and CATO chooses the tools. Checking the method, package versions, assumptions, reference data, code, and interpretation is still up to you.

How Code Runs

One conversation, one session

CATO keeps one working directory and shell per conversation, so files and state carry between messages.

Long commands

A long command keeps running if you leave the page, and CATO carries on when it ends. The Shell view shows each command, its output, and a Stop button.

Code runs in an isolated sandbox, with limits on network access, computing resources, and running time. Files stay on the project computer. A program still running when the computer shuts down ends, so ask CATO to save anything you need.

Python and R

CATO can write scripts and notebooks in either language, run them, and save tables, figures, and results. Common biomedical and statistical R packages are available alongside the Python ones.

Name the language when it matters: write “Run this analysis in Python” or “Run this analysis in R”, along with any package, version, or output format you need.

  • Ask CATO to check that a package and version are present before you plan a long workflow around it.
  • Ask it to set random seeds where results depend on them.
  • Save intermediate tables as files rather than relying on what the session holds in memory.
  • Download the notebook, scripts, package versions, and outputs needed to reproduce a result.

What CATO Can Work With

Core data science and machine learning

pandas, NumPy, SciPy, statsmodels, scikit-learn, gradient boosting, model diagnostics, and visualization.

Genomics and single-cell

Sequence and variant formats, BioPython, pysam, scanpy, anndata, differential expression, clustering, dimensionality reduction, and pathway analysis.

Proteomics and mass spectrometry

mzML and MGF files, pyOpenMS, pyteomics, quality control, normalization, peptide and protein summaries, and downstream statistics.

Medical imaging and signals

DICOM and NIfTI files, pydicom, nibabel, SimpleITK, image metadata, segmentation, and signal processing.

Molecular and chemical analysis

RDKit, structure files, molecular descriptors, similarity and substructure searches, and interactive 3D views.

Causal, survival, and longitudinal analysis

Survival models, time-to-event diagnostics, mixed and longitudinal models, causal diagrams, effect estimation, sensitivity analysis, and time series.

Networks, pathways, and ontologies

Graph analysis, enrichment, gene sets, pathway databases, community structure, centrality, and biological identifier lookups.

Literature and research databases

Scholarly search, clinical-trial and patent lookups, open-access retrieval, and evidence synthesis.

These are examples, not a promise that every package or outside service is available. Ask CATO to check.

A Reliable Analysis Workflow

  1. Define the question. Name the population, variables, units, and the decision the result will inform.
  2. Look at the data first. Profile the columns, missing values, duplicates, ranges, and coding.
  3. Approve a plan. Ask for a plan first for exclusions, transformations, models, multiple testing, or evidence criteria.
  4. Save your work. Save code, parameters, package versions, logs, and results as files.
  5. Check the model. Look at assumptions, convergence, residuals, robustness, leakage, and uncertainty.
  6. Look at the output. Check every table, figure, and structure against the source data.
  7. Reproduce it. Re-run important results from a clean state and compare.

Notebooks

Notebooks open in the Workspace with their code and output, and each version is kept. You do not run cells yourself: ask for the change and read the new version.

A notebook can depend on session state you cannot see. If it only works when its cells run in one order, it is not reproducible. Ask for a clean run from the top before you treat it as final.

Figures

Figures are saved as files with their versions, not only shown as chat images. You can edit an SVG figure without losing the original, and 3D and other scientific views keep their own controls.

  • Check plotted values, transformations, scales, units, and denominators.
  • Use versions to keep analysis changes separate from cosmetic edits.
  • Check labels, clipping, legends, contrast, and the downloaded file.
  • Keep the data and code behind a figure with the figure.

Where a Result Came From

A run can produce datasets, scripts, notebooks, figures, reports, or manuscripts. Each is saved in the project with its versions, and Provenance, in a file's menu, shows how each version was made.

Before you publish, be able to trace each claim: source file, code, package versions, result, then the figure or table that shows it.

Outside Databases

Lookups depend on the outside database being available. Record accession numbers, database versions or access dates, queries, and URLs. A failed lookup does not mean the record does not exist.

What CATO Does Not Replace

  • CATO helps with research analysis. It does not replace clinical judgment, a validated diagnostic device, or regulatory review.
  • Generated code can contain statistical, computational, or security errors. Read it before you reuse it.
  • Do not put passwords or keys in code or notebooks.
  • Check software licenses and data-use restrictions before you redistribute a result or environment.
  • Take high-stakes findings to the right statistician, domain expert, privacy officer, or safety reviewer.