TrendBio

#22 · ML / AI methods

Causal and mediation inference for high-dimensional omics with uncertainty control

1 preprint · 4 published
Items per day, last 8 weekstoday: 5

The common thread is inference on high-dimensional omics that targets causal or conditional structure with explicit uncertainty control. Mendelian randomization work, dmIVW, handles uncorrelated and correlated pleiotropy, while TL-HDMR applies transfer learning across ancestries to separate shared from ethnic-specific metabolite causes of stroke. Mediation methods extend to heterogeneous effects with type I error control in M-high-learner and to gene-set methylation mediation in gHDMA. Cooperative learning with penalized linear mixed-effects models, CL-pLMM, addresses clustered multiview data, and CICIV ranks causal transcriptomic drivers of breast cancer progression by combining PC-simple feature selection with conditional instrumental variable representation learning. What is new is the pairing of uncertainty quantification with heterogeneity, ancestry, clustering, and multiview structure rather than association alone.

Published papers

Preprints

Topic history

DateRank
#22
#28
#34
#22
#11
#27
#36
#38
#40
#29
#36
#27
#4
#13
#10
#11
#6
#10
#4
#6
#17
#13
#4
#23
#7
#38
#35
#5
#31
#38
#28
#33
#19
#39
#2
#26
#15
#19
#9
#18
#16
#16
#2
#3
#31
#30
#5
#13
#8
#13
#16
#9
#30
#40
#21
#14
#35
#3
#11
#12
#12
#19
#31
#13
#24
#10
#37
#32
#26
#12
#36
#37

Sponsors