Biological imaging
Evaluate ML/DL methods for cell segmentation and measurable image features.
Connect the image, the molecule and the experiment.
A proposed computational biology module would bridge wet-lab experiments and in silico analysis, using biological images and multi-omics data to develop testable hypotheses.
AI within the scientific workflow
The development proposal explores artificial intelligence (AI), machine learning (ML) and deep learning (DL). The capabilities below require suitable datasets, implementation and validation.
Proposed moduleOmics and multi-omics
Explore molecular profiles and compatible datasets to identify relationships worth testing.
Experimental feedback
Use computational findings to frame comparisons, then refine hypotheses from laboratory results.
From observation to the next experiment
The intended workflow keeps computational interpretation connected to the biological question.
Capture
Record images, molecular measurements and experimental conditions.
Analyse
Check data quality and compare image features with relevant molecular profiles.
Form a hypothesis
Prioritise a testable relationship, including alternative explanations.
Test and learn
Design a controlled experiment and use its results to update the analysis.
Methods to evaluate
Method selection would follow the data type, sample design and question.
Read more
Image analysis
Compare candidate segmentation and feature-extraction methods with expert annotations.
Molecular analysis
Examine RNA expression and other available molecular measurements with appropriate controls and normalisation.
Data integration
Explore compatible imaging and multi-omics measurements while accounting for batch effects and missing data.
Scientific review
Interpret patterns with biological context and uncertainty; validate proposed relationships experimentally.
What would establish a capability
Each implemented analysis would need a defined task, comparison and validation record before a performance claim.
Proposed evaluationRead more
Data provenance
Document data sources, permissions, sample handling and annotation quality.
Independent evaluation
Use suitable baselines and held-out samples, separating donors and batches where needed to prevent data leakage.
Reproducibility
Record model versions, analysis settings and failure cases.
Biological follow-through
Test whether a computational finding holds in the relevant experimental system.
Selected sources
External method study
Cellpose: a generalist algorithm for cellular segmentationStringer et al., 2021. A deep-learning segmentation method; cited as methodological context, not an adopted or validated Rnevox capability.
External method study
MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell dataArgelaguet et al., 2020. Multi-modal data integration research; an example of methods to evaluate, not a Rnevox implementation.