AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Bayesian multi-domain learning for cancer subtype discovery from next-generation sequencing count data

Ehsan Hajiramezanali, Siamak Zamani Dadaneh, Alireza Karbalayghareh, Mingyuan Zhou, Xiaoning Qian

NeurIPS 2018 · Probabilistic and scientific modeling › Molecules and proteins · approach: Statistical analysis of ML practice · 30 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use domain-specific factor selectors with the beta-Bernoulli process instead of assigning every factor to every domain (ablation)BMDL versus hGNBP · Lung cancer, low-related domains, Ns=25 and Ns=100accuracy (%): 72.94 ± 1.70 -> 78.85 ± 4.55; 74.55 ± 8.84 -> 78.10 ± 5.65Improved
Use domain-specific factor selectors with the beta-Bernoulli process instead of assigning every factor to every domain (ablation)BMDL versus hGNBP · Lung cancer, highly-related domains, Ns=25 and Ns=100accuracy (%): 74.13 ± 7.07 -> 78.46 ± 5.97; 77.61 ± 3.54 -> 81.49 ± 5.12Improved
Integrate source-domain data with domain-adaptive factors rather than learn factors from target samples onlyBMDL versus hGNBP-NBFA · Lung cancer, highly-related and low-relatedaccuracy (%): 73.38 ± 7.29 -> 78.46 ± 5.97; 73.38 ± 7.29 -> 78.85 ± 4.55Improved
Model sample-specific negative-binomial dispersion instead of fixing the NB probability parameter (ablation)HDP-NBFA versus NB-HDP · Lung cancer, highly-related domainsaccuracy (%): 55.22 ± 3.69 -> 63.48 ± 1.23; 56.52 ± 4.61 -> 65.65 ± 4.22Improved
Use hierarchical gamma-process factor weights rather than hierarchical Dirichlet-process weightshGNBP versus HDP-NBFA · Lung cancer, low-related domainsaccuracy (%): 54.89 ± 7.38 -> 72.94 ± 1.70; 51.83 ± 8.32 -> 74.55 ± 8.84Improved
Add more source samples under a forced shared-factor structureHDP-NBFA · Lung cancer, low-related domainsaccuracy (%): 54.89 ± 7.38 -> 51.83 ± 8.32Hurt
Add more source samples under a forced shared-factor structureNB-HDP · Lung cancer, low-related domainsaccuracy (%): 54.57 ± 7.73 -> 53.83 ± 7.79Hurt
Increase target-domain sample countBMDL and hGNBP-NBFA · Synthetic dataclassification errorImproved
Increase the number of shared latent factors between domainsBMDL · Synthetic dataclassification errorImproved
Add source-domain data through BMDL factor sharingBMDL versus target-only hGNBP-NBFA · Synthetic data, varying domain relevanceclassification errorImproved
Add another low-related source domain when the highly-related domain already has 100 samplesBMDL · Lung cancer, three domainsaccuracy (%)No clear effect
Add a highly-related source domain with 100 samples to 100 low-related source samplesBMDL · Lung cancer, three domainsaccuracy (%): around 4%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy