DeepProfile: Deep learning of cancer molecular profiles for precision medicine
Ayse Berceste Dincer, Safiye Celik, Naozumi HiranumaiD, Su‐In LeeiD
Abstract
We present the DeepProfile framework, which learns a variational autoencoder (VAE) network from thousands of publicly available gene expression samples and uses this network to encode a low-dimensional representation (LDR) to predict complex disease phenotypes. To our knowledge, DeepProfile is the first attempt to use deep learning to extract a feature representation from a vast quantity of unlabeled (i.e, lacking phenotype information) expression samples that are not incorporated into the prediction problem. We use Deep-Profile to predict acute myeloid leukemia patients in vitro responses to 160 chemotherapy drugs. We show that, when compared to the original features (i.e., expression levels) and LDRs from two commonly used dimensionality reduction methods, DeepProfile: (1) better predicts complex phenotypes, (2) better captures known functional gene groups, and (3) better reconstructs the input data. We show that DeepProfile is generalizable to other diseases and phenotypes by using it to predict ovarian cancer patients tumor invasion patterns and breast cancer patients disease subtypes.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 54
- Organism
- Assay
- Research Focus
- Approach
Abstract summary
Claude Opus 5.5 · xhigh effort · 2 Nov 2026 · from the abstract
DeepProfile trains a variational autoencoder on thousands of unlabeled gene expression samples to learn low-dimensional representations that improve prediction of acute myeloid leukemia drug responses and cancer phenotypes.
DeepProfile trains a variational autoencoder (VAE) on thousands of publicly available unlabeled gene expression samples to encode low-dimensional representations (LDRs) for predicting complex disease phenotypes. The authors describe this as the first use of deep learning to extract feature representations from large volumes of unlabeled expression samples not included in the downstream prediction problem.
Key results
SuggestedDeepProfile outperforms raw gene expression levels and two standard dimensionality reduction methods in predicting in vitro responses to 160 chemotherapy drugs in acute myeloid leukemia patients, capturing functional gene groups, and reconstructing input expression data.
Evaluated on acute myeloid leukemia patient in vitro responses to 160 chemotherapy drugs against raw expression features and two dimensionality reduction methods; numerical performance metrics are not provided in the abstract. · Abstract
SuggestedRepresentations learned by DeepProfile generalize beyond acute myeloid leukemia to predict tumor invasion patterns in ovarian cancer patients and disease subtypes in breast cancer patients.
Applied to ovarian cancer tumor invasion pattern prediction and breast cancer subtype classification; quantitative accuracy values and cohort sizes are not stated in the abstract. · Abstract
Methods, data and reproducibility
The authors trained a variational autoencoder (VAE) on thousands of publicly available unlabeled gene expression samples to encode a low-dimensional representation (LDR), then used these features to predict acute myeloid leukemia responses to 160 chemotherapy drugs, ovarian cancer invasion patterns, and breast cancer subtypes.
- Publicly available unlabeled gene expression samples · n = thousands of samples
- Acute myeloid leukemia patient gene expression and in vitro responses to 160 chemotherapy drugs · human · n = 160 chemotherapy drugs
- Ovarian cancer patient samples (tumor invasion patterns) and breast cancer patient samples (disease subtypes) · human
- Methods & tools
- DeepProfile
- Reproducibility gaps
- Abstract only; no code or dataset links are provided.
Limitations (1)
- The abstract does not report quantitative prediction metrics, exact sample sizes, or the names of the two baseline dimensionality reduction methods. · noted in this summary
Who should read:Read this if you want to pretrain variational autoencoders on unlabeled gene expression compendia to predict patient drug responses or clinical cancer phenotypes.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- A deep profile of gene expression across 18 human cancers
Wei Qiu, Ayse B. Dincer, … Su‐In Lee·bioRxiv·Mar 17, 2024
- Extracting a Biologically Relevant Latent Space from Cancer Transcriptomes with Variational Autoencoders
Gregory P. Way, Casey S. Greene·Pacific Symposium on Biocomputing (PSB)Peer-reviewed·Nov 17, 2017
- Predicting chemotherapy response using a variational autoencoder approach
Qi Wei, Stephen A. Ramsey·BMC BioinformaticsPeer-reviewed·Sep 22, 2021
- Deep representation learning for temporal inference in cancer omics: a systematic review
Guillermo Prol-Castelo, Davide Cirillo, Alfonso Valencia·bioRxiv·May 29, 2025
- JavaDL: a Java-based Deep Learning Tool to Predict Drug Responses
Beibei Huang, Lon Wolf R. Fong, … Shuxing Zhang·bioRxiv·May 4, 2020
- Deep Learning and Association Rule Mining for Predicting Drug Response in Cancer. A Personalised Medicine Approach.
Konstantinos N. Vougas, Magdalena Krochmal, … Vassilis G. Gorgoulis·bioRxiv·May 9, 2017