Unsupervised Domain Adaptation for Deconvolution of Spatial Transcriptomics Spots
Bibliographic record
Abstract
Interest in local intercellular communication within tissues in fields such developmental biology, neuroscience, and oncology have motivated the development of spatially resolved transcriptomics technologies.These technologies come with trade-offs; for in situ capture methods such as 10x Genomics Visium, for example, the gene expression captured from a single spot may come from multiple cells, and thus computational methods for integrating single-cell data to deconvolve cell type proportions have gained interest.To date, these have required samplematched datasets from each modality, whereas in most cases these are not available.Taking inspiration from and building on CellDART, a method that uses domain adaptation to address this problem by treating reference scRNA-seq data as source domain data, we further investigate the performance of 3 domain adaptation methods (ADDA, DANN, and Deep CORAL) and a Bayesian method (RCTD) on 3 different pairs of source and target datasets (dlPFC, PDAC, and a "gold standard" mouse cortex dataset).We used metrics across 3 levels of evaluation: (a) performance on source domain scRNA-seq derived "pseudo-spots", (b) performance in mapping inputs from both domains to a common distribution, and (c) performance on real spatial transcriptomics spots.By integrating these 3 datasets and 3 levels of metrics, we ensured that our results were robust and broadly applicable.Upon performing hyperparameter tuning for all three models across all three sets, and evaluating the models, we were unable to conclusively determine whether any method performed better or worse.We found that CellDART remained a good performer, that it was difficult to train DANN, and that ADDA's performance is limited by not remaining consistent for specific samples.Future directions include improving on CellDART, integrate cycle-consistent loss for ADDA, and implementing unsupervised domain adaptation validation methods.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.001 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".