Högskolan i Skövde

his.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • apa-cv
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
How Deep Multimodal Fusion Affects Uncertainty and Calibration Using Transcriptomics data
University of Skövde, School of Bioscience.
2024 (English)Independent thesis Advanced level (degree of Master (One Year)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Deep neural networks (DNNs) have become integral to predictive modelling in high-dimensional datasets, such as omics data, yet their overconfidence in predictions poses significant challenges for safety-critical tasks like medical diagnostics. This study investigates whether multimodal fusion of omics data, specifically RNA-Seq and miRNA-Seq, can improve the uncertainty quantification (UQ) and calibration of DNNs when predicting overall survival in cancer patients. Using data from The Cancer Genome Atlas (TCGA) for Ovarian Serous Cystadenocarcinoma (TCGA-OV) and Breast Invasive Carcinoma (TCGA-BRCA), a multimodal DNN architecture with separate feature extraction branches for each modality implemented, followed by concatenation and classification. Predictive uncertainty was quantified using Monte Carlo Dropout (MCD) with metrics such as mean predictive variance, uncertainty ratio, and entropy, while calibration was assessed using Adaptive Calibration Error. Performance evaluation, guided by Balanced Accuracy (BA) and nested cross-validation, revealed mixed results: the multimodal model showed higher predictive variance (i.e. higher uncertainty awareness) and mean BA for TCGA-BRCA but performed worse for TCGA-OV. Although predictive variance for the BRCA dataset carried statistical significance, the lack of significance across other key metrics limited definitive conclusions. These findings suggest that multimodal fusion may increase uncertainty under specific conditions but highlight the need for more sophisticated architectures and comprehensive analyses to validate these effects.

Place, publisher, year, edition, pages
2024. , p. 27
National Category
Bioinformatics (Computational Biology)
Identifiers
URN: urn:nbn:se:his:diva-24912OAI: oai:DiVA.org:his-24912DiVA, id: diva2:1938634
Subject / course
Bioinformatics
Educational program
Bioinformatics - Master’s Programme
Supervisors
Examiners
Available from: 2025-02-19 Created: 2025-02-19 Last updated: 2025-09-29Bibliographically approved

Open Access in DiVA

fulltext(965 kB)165 downloads
File information
File name FULLTEXT01.pdfFile size 965 kBChecksum SHA-512
ac90ba181469118c334aa61ac4d6423025445b26791d307da2b5d6cc8e07a84073e4811a30f4b7404900c4c54d08134e48373e74816271bbf8e3a2f76b332d6f
Type fulltextMimetype application/pdf

By organisation
School of Bioscience
Bioinformatics (Computational Biology)

Search outside of DiVA

GoogleGoogle Scholar
Total: 165 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 311 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • apa-cv
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf