2 of 16

What is the CQS?

An SRI service, supported by Exposures Provider
Conceptualized and designed by the Translator Clinical Data Committee (TCDC)
Provides ARA-like capabilities to support a “Conservative Ingest” paradigm (link) via a centralized source with known conditions and provenance tracing
Supports manually curated and SMuRF- and SME-evaluated workflows
Workflow paths are structured as TRAPI queries; Workflow Runner executes queries
Currently supports three workflow paths, each responsive to the inferred MVP1 query: What drugs may treat disease X?

Not restricted to MVP1, but rather capable of supporting any inferred queries
Designed to support curated workflow queries developed by any team, working group, committee, or even external contributors
Prompted the need for Translator Release Process Guidelines

https://github.com/NCATSTranslator/Clinical-Data-Committee-Tracking-Voting/tree/main/GetCreative()_DrugDiscoveryRepurposing_RarePulmonaryDisease

(Practical) benefits of leveraging the CQS as an ARA-like service, in addition to those listed above, include::

Allowing SMuRFs/ SMEs or teams / WGs / Committees to contribute their own curated inferences to inferred MVP queries
Allowing KPs such as OpenPredict or Multiomics to avoid dealing with ARA functions such as ARS registration, merging, scoring, normalizing, adding OmniCorp literature co-occurrence AuxGraphs, etc., thus minimizing the burden on KPs and avoiding redundancy
Allowing KPs to contribute their own “special sauce” to inferred MVP queries, without having to rely on other ARAs
KPs need to have greater input on inferences over their data
CQS provides a central, single source of inference rules
KPs don’t want to make aux graphs
CQS supports multiple rules
CQS supports TAQA testing

Challenges:

1) KPs need greater influence on what inferences are made on their data

2) We have a lot of disagreement on what should constitute an inference and we have no place to discuss.disagree.collaborate.deduplicate

3) KPs don’t want to make aux graphs

4) lots of examples of one-hop inference / predication (~) /deterministic rule or outcome - GOCAM, expression, variants and loss of function

Higher-level issues and implementation (CQS)

One limitation of the CQS is that KP teams may want to expose the inferred treats edges as "knowledge" in general that could be used in other non-MVP1 Translator queries by the ARAs (i.e., an ARA answers a different question where the treats edge is just one hop within a multihop query). However, The same is true for all other ARA-predicted treats edges, unless there is something 'special' about these one-hop treats predictions that warrants them (but not other predicted treats edges) being persistently stored and accessible for subsequent reasoning.

tracker - https://docs.google.com/spreadsheets/d/11faQ3IA56hW1XdYF8cTbWu7aeYC9xSvBW06hsWRSu-Y/edit?usp=sharing

3 of 16

CQS Driving Use Cases

Leverage clinical KP “knowledge”
Leverage CHP’s gene-tissue enrichment functionality
Leverage OpenPredict’s RWE-based predictive algorithms on drugs that may treat disease X
Leverage Multiomics Clinical Trial Providers’ SMuRF assertions on drugs that may treat disease X

… others likely

4 of 16

High-level Overview of CQS Architecture

ARS

ARA

CQS

TRAPI inferred request message

Decision point: biolink:treats, inferred:true, & one or more matched CQS q-graph templates:

if no -> ignore;

if yes -> send matched q-graph TRAPI message to WFR

Responses returned to CQS and then merged, scored, and normalized

TRAPI response message

ARAs

KPs

WFR

For details, see https://github.com/NCATSTranslator/Clinical-Data-Committee-Tracking-Voting/issues/17

Purpose is to support ARA-like “Conservative Ingest” (link) via a centralized source

What drug may treat disease X?

Drug Y may treat disease X

AuxGraphs = MVP Templates 1, 2, 3 …

5 of 16

MVP1 Workflow Path A: Conceptual Design

Chemical Entity (n3)

Disease (n0)

(Input CURIE)

Gene (n2)

Chemical Entity (n1)

allowlist: cohd, icees-kg, multiomics-ehr-risk

(currently does not work)

Leverages clinical “knowledge” in first hop and text-mined assertions / molecular “knowledge” in second and third hop
Identifies drugs/chemicals that are associated with a given disease in RWE, then finds drugs/chemicals that act on the same genes that are targeted by the drugs/chemicals that are associated with a given disease in RWE (i.e., first hop)

e0:biolink: correlated_with, associated_with_likelihood_of

e1:biolink:affects, directly_physically_interacts_with

allowlist: text-mining-provider-targeted, molepro

e3:biolink: contributes_to, associated_with, gene_associated_with_condition

e2:biolink:affects, directly_physically_interacts_with

allowlist: text-mining-provider-targeted, molepro

CQS inference:

n3 ChemicalEntity - biolink:treats - n0 Disease

based on Path A AuxGraph

Caveat re Multiomics EHR Risk Provider: https://github.com/TranslatorSRI/CQS/issues/1

ETA for dev deployment: 11/22

6 of 16

MVP1 Workflow Path B: Conceptual Design

Drug / Chemical Entity (n3)

Disease (n0)

(Input CURIE)

Gene (n2)

Tissues (n1)

allowlist: chp

chp:gene-tissue enrichment

Leverages CHP’s gene-tissue enrichment functionality
Identifies drugs/chemicals that are associated with genes enriched in tissues of relevance to a given disease

biolink:located_in

biolink:expresses

biolink:affects

CQS inference:

n3 ChemicalEntity - biolink:treats - n0 Disease

based on Path B AuxGraph

ETA for dev deployment: 11/22

7 of 16

MVP1 Workflow Path E: Conceptual Design

Drug / Chemical Entity (n1)

Disease (n0)

(Input CURIE)

biolink:treats

Leverages OpenPredict’s RWE-based algorithmic predictions on drugs that may treat a given disease
Designed as a simple one-hop

allowlist: open-predict

CQS inference*:

n1 ChemicalEntity - biolink:treats - n0 Disease

based on Path E AuxGraph

*In this case, CQS is “agreeing” with OpenPredict’s “treats” inference

ETA for dev deployment: 11/22

8 of 16

Proposed MVP1 Workflow Path F: Conceptual Design

Drug / Chemical Entity (n1)

Disease (n0)

(Input CURIE)

biolink:in_clinical_trials_for

Leverages Multiomics Clinical Trials Providers’ SMuRF assertions on drugs that may treat a given disease
Designed as a simple one-hop

allowlist: multiomics-clinical-trials

CQS inference:

n1 ChemicalEntity - biolink:treats - n0 Disease

based on Path F AuxGraph

Caveat re Multiomics EHR Risk Provider: https://github.com/TranslatorSRI/CQS/issues/1

Proposed for next release cycle

9 of 16

Proposed MVP2 Workflow Path A: Conceptual Design

Leverages CAM-KP’s Causal Activity Model (CAM) assertions to provide mechanistic insights into real-world clinical observations
Designed as a simple two-hop
The demonstration query asks for chemical entities associated with disease diagnoses in the real world and then asks for genes or gene products that are affected by those chemicals. This example may seem simplistic and not terribly novel; however, cam-kp contains CAM representations that provide mini-KGs or networks of annotations that together provide a more complete model of biological function than the separate annotations (e.g., a network of how different gene products work together in a biological pathway).

Chemical Entity (n1)

Disease (n0)

(Input CURIE)

biolink:correlated_with, has_increased_likelihood_of

allowlist: cohd, icees-kg, multiomics-ehr-risk-kp

Gene (n2)

biolink:affects

allowlist: cam-kp

CQS inference:

n1 ChemicalEntity - biolink:affects - n2 Gene in context of n0 Disease

based on Path A AuxGraph / cam-kp AuxGraph

Proposed for next release cycle

10 of 16

Proposed MVP2 Workflow Path A: Example

ChemicalEntity (n1)

NamedThing (n2)

biolink:affects (qualified)

allowlist: cam-kp

CQS inference:

n1 ChemicalEntity - biolink:affects - n2 BiologicalProcessOrActivity

based on cam-kp AuxGraph

CAM curated workflow path

ChemicalEntity (n1): nitric oxide

AOP 149 (https://aopwiki.org/aops/149). It is still pretty busy but hopefully is clear enough. Base process is oxidative stress -> Nitric oxide synthesis -> nitric oxide -> vasodilation -> increased systematic vascular resistance -> hypertension and that there are other processes that feed into that base process along the way. Note that there are four CAM-ified AOPs related to nitric oxide; the example here represents one of them

Title: Peptide Oxidation Leading to Hypertension

Description from the wiki:

Here we present the supporting information on an AOP describing how vascular endothelial peptide oxidation leads to hypertension via perturbation of endothelial nitric oxide (NO) bioavailability. The molecular initiating event is oxidation of amino acid (AA) residues on critical peptides of the NO pathway, notably protein kinase B (AKT), guanosine triphosphate cyclohydrolase-1 (GTPCH-1), endothelial nitric oxide synthase (eNOS), and also the cellular ROS scavenger; glutathione. Oxidation of the enzymic components of the pathway lead to reduced expression of the phosphorylated proteins, and protein loss via proteasomal degradation. Oxidation of reduced glutathione to GSSG promotes bonding of GSSG to critical AA residues on eNOS, and the reduced expression of GTPCH-1 reduces bioavailability of tetrahydrobiopterin (BH4), both of which lead to uncoupling of eNOS (reduced NO production, increased superoxide production). The combination of these molecular events lead to reduced bioavailabilty of NO, which in turn reduces the potential for vasodilation and shifts the balance of vascular tone towards vasoconstriction. Repeated perturbation of this pathway via chronic exposure to toxicants, ultimately increases vascular resistance and contributes towards the development of hypertension.

11 of 16

CAM KP edges

1 biolink:colocalizes_with

23 biolink:has_attribute

45 biolink:phenotype_of

58 biolink:coexists_with

386 biolink:temporally_related_to

17132 biolink:acts_upstream_of_negative_effect

17238 biolink:acts_upstream_of_or_within_negative_effect

21249 biolink:is_output_of

24237 biolink:has_output

26863 biolink:directly_physically_interacts_with

32123 biolink:active_in

35980 biolink:location_of

35994 biolink:located_in

40838 biolink:has_part

40838 biolink:part_of

46390 biolink:capable_of

46390 biolink:enabled_by

46390 biolink:enables

65802 biolink:interacts_with

70691 biolink:acts_upstream_of_positive_effect

81286 biolink:acts_upstream_of_or_within_positive_effect

110243 biolink:overlaps

156277 biolink:acts_upstream_of_or_within

158822 biolink:contains_process

160662 biolink:actively_involved_in

166062 biolink:occurs_in

205323 biolink:preceded_by

205323 biolink:precedes

319823 biolink:regulates

483984 biolink:caused_by

484974 biolink:causes

675221 biolink:has_input

675221 biolink:is_input_of

3754751 biolink:participates_in

3782566 biolink:has_participant

5154895 biolink:affected_by

5160466 biolink:affects

12 of 16

Questions? Concerns?

13 of 16

Proposed MVP2 Workflow Path A: Example

MolecularActivity (n1)

MolecularActivity (n2)

biolink:affects (qualified)

allowlist: cam-kp

CQS inference:

n1 MolecularActivity - biolink:affects - n2 MolecularActivity

based on cam-kp AuxGraph

CAM curated workflow path

14 of 16

MVP1: What drugs may treat disease X?

biolink:cooccurs_in_literature_with

Branch CHP query refinement, ops

(Greg H.)

Tissue?

Gene?

Drug, Small Molecule?

Edge support from Text Miner

(Bill B.)

biolink:has_real_world_evidence_of_association_with

https://github.com/NCATSTranslator/Clinical-Data-Committee-Tracking-Voting/tree/main/GetCreative()_DrugDiscoveryRepurposing_RarePulmonaryDisease/

Edge support from CAM KP

(Gaurav V., Kara F.)

Edge support from CAM KP

(Gaurav V., Kara F.)

Branch query refinement, ops

(Kara F., Casey T., David K.)

COHD, ICEES-KG (Kara F., Casey T. Max W.),

Multiomics EHR Risk Provider to be added

(Basa B. Kamileh N.)

DiseaseOrPhenotypicFeature?

Gene?

ChemicalEntity?

gene_associated_with_condition

RWE

physically_interacts_with

Branch query refinement, ops

(Kara F., Casey T., David K.)

Path A

Path B

Path C

(Workflow B)

Genetics Provider

(Marc D.)

MolePro

(Vlado D.)

What drugs may treat rare pulmonary disease?

15 of 16

MVP1 Workflow Path A: Conceptual Design

Chemical** (n3)

Disease (n0)

(Input CURIE)

Gene (n2)

Chemical**

(n1)

allowlist: cohd, icees-kg, multiomics-ehr-risk

(currently does not work)

Leverages clinical KP “knowledge” in first hop
Identifies drugs/chemicals that are associated with a given disease in RWE
Finds drugs/chemicals that act on the same genes that are targeted by the drugs/chemicals that are associated with a given disease in RWE (i.e., first hop)

e0: biolink: correlated_with, associated_with_likelihood_of

e1: biolink: affects (qualified)*

*causes increased activity_or_abundance/expression/secretion, decreased degradation

**MolecularEntity, EnvironmentalExposure

e2: biolink: affects (qualified)*

e3: biolink: contributes_to, associated_with, gene_associated_with_condition

16 of 16

MVP1 Workflow Path A.1: Conceptual Design

Chemical Entity (n3)

Disease (n0)

(Input CURIE)

Gene (n2)

Chemical Entity (n1)

allowlist: cohd, icees-kg, multiomics-ehr-risk

(currently does not work)

e0:biolink: correlated_with, associated_with_likelihood_of

e1:biolink: directly_physically_interacts_with

e2:biolink: directly_physically_interacts_with

e3:biolink: contributes_to, associated_with, gene_associated_with_condition

Path A.1 AuxGraph =

returned answer subgraphs + EPC

Leverages clinical “knowledge” in first hop and text-mined assertions / molecular “knowledge” in second and third hop
Identifies drugs/chemicals that are associated with a given disease in RWE, then finds drugs/chemicals that act on the same genes that are targeted by the drugs/chemicals that are associated with a given disease in RWE (i.e., first hop)