Researchers at the University of Southern California have developed a powerful new tool called DeepDeconUQ that improves how scientists estimate the percentage of cancer cells in tissue samples using RNA sequencing data. This advancement is important for doctors and researchers who rely on accurate cancer cell measurements to make informed decisions about diagnosis, treatment, and disease progression.

Traditionally, many tools provide only a single best guess of how many cancer cells are present in a tissue sample. However, these estimates don’t tell you how certain or uncertain that guess is, which can be a serious limitation in medical settings. The team at USC approached this challenge by designing a deep learning model that not only estimates cancer cell proportions but also shows how confident the model is in its predictions.

Overview of DeepDeconUQ

A: Constructing simulated bulk RNA-seq samples with different fractions of malignant cells. p is the fraction of malignant cells in a simulated bulk sample. B: Model structure used to train DeepDeconUQ. It consists of four fully connected layers with dropout layers. Seventy percent of the simulated data are used for training. The output is two quantile functions at a given significance level . C: Conformity scores are calculated on the remaining 30% of the simulated dataset. D: Estimating the prediction interval of malignant cells from a real bulk sample. The trained model is used to calculate the lower and upper bounds, and the conformity scores are used to adjust the quantiles, which finally outputs the prediction interval .

DeepDeconUQ uses bulk RNA sequencing data and integrates it with single-cell RNA sequencing data to learn the patterns of gene expression that distinguish cancer cells from normal cells. It applies a type of machine learning called quantile regression to generate a range, rather than just one number, that captures the likely percentage of malignant cells in a sample. These ranges are then calibrated to make sure they are statistically reliable and as narrow as possible, meaning they are both accurate and precise.

The model was tested on both simulated and real-world cancer datasets and showed superior performance compared to other methods. It also held up well when the input gene expression data were slightly altered, showing strong robustness. This combination of accuracy, reliability, and resilience makes DeepDeconUQ a promising tool for both clinical and research applications.

Availability – The DeepDeconUQ method is publicly accessible at https://github.com/jiaweih14/DeepDeconUQ.