BenevolentAl Therapeutics Pipeline and Triage
Ulcerative Colitis Example: molecular-signature
detection linked to outcomes
Z
Tm
(SOW)
m
k=1...K
n=1...N
m=1...M
xm
nd
d=1...D
Transcriptomes
organised into
latent structure
PATIENT-LEVEL
DATA
Dataset of IBD
patient samples
(N = 105 UC/Crohn's
patient samples)
Generative ML models
Inflammatory marker
response from suite
of models
Genes
Patients
PATIENT SAMPLES CLUSTERED
CD56dim NK
Pathway marker Y
M1 macrophage
11
0.20-
R = 0.62
FDR<0.001
10-
0.2-
0.15-
0.10-
0.1-
0.05-
R=0.65
FDR<0.001
R=-0.74
FDR<0.001
0.0
0.00
ૐ
neutrophil
Target X
0.3-
0.2-
•
0.1-
R=0.63
FDR<0.001
6- R= -0.63
FDR<0.001
0.0
-90
-60
-90
-60
Latent 75
disease state
CD
Ctr
UC
Entity Explorer
IBD Latent List (modified)
Disease program: IBD Demo
Data release: 1-27 (latest)
How much do shortlisted entities
overlap?
Number of edges displayed: 1315
Structured data relationships
High-confidence literature-derive
☐
ed relationships
+
Q Search
Explore
Visualise
Shortlists
Data-derived mechanism entities are integrated
into the Knowledge Graph and connected to
enriched biological pathways and relevant
biomedical entities (e.g. diseases, tissues, targets)
ML models recapture specific subgroups with key inflammatory markers (IL1 and TNFa signalling) and immune
cells (M1 macrophages and neutrophils) and uncover mechanistic areas to explore further
Al
Benevolent 27View entire presentation