About
I am Professor of Informatics and Computing at Indiana University Luddy School, where I direct RedLab and lead a group researching representing and reasoning for data that is messy, incomplete, and spread across sources. These methods are applied in drug discovery, healthcare, emergency response, cybersecurity and defense.
I trained as a computer scientist and spent the first part of my career in the pharmaceutical industry, leading scientific computing teams in computational chemistry and molecular modeling. I came to Indiana University in 2005 from that industry background, which still shapes how I work: I am equally interested in whether something functions in the field and whether it is elegant in theory.
For over ten years I practised as a volunteer EMT and emergency manager, which gave me a new lens on technology and data in critical decision making. The gap between the data that exists and the decision that has to be made is what most of my work is about.
I love creating and exploring new ideas and possibilities. I still write code, I am a persistent tinkerer with new tools, and I read widely outside my field. I write about AI and its wider consequences at Prior Shift. Originally from the UK; based in Bloomington, Indiana.
Research scope
My research concerns acquiring, building and reasoning over data in domains where data acquisition is difficult, data is incomplete and heterogeneous, and decisions are consequential. Methods include graph, semantic and topological representation, machine learning, embeddings and language models and human-computer interaction design frameworks that put structure in front of the person who has to act.
Application Areas
The mathematical methods apply to any domain where the data is complex, heterogeneous and incomplete. They were developed against the most complex system we know of - the human body - and are now being applied in areas including situational awareness and cybersecurity.
Grounded in
Where the methods were built and where they continue to impact: cheminformatics, bioinformatics, chemogenomics, and link prediction over sparse biological evidence — most recently yielding experimentally confirmed drug candidates for tuberculosis.
Current focus
First responders, disaster responders, and emergency managers have to make decisions in complex, adverse, and data overload contexts. My focus is on filling data gaps, fusing heterogeneous real-time streams, and surfacing the signal in an actionable way in high stress environments.
Extending to
Increasingly decisions have to be made across domains and with overlapping data and expertise structures, for example, in cyber-physical systems. These same methods can be applied at higher or different abstraction levels across domains, maybe even to enable decisions that go beyond human capacity to acquire and fuse data.
Research topics
For full publication list see my Google Scholar profile.
These two older papers show how a domain (in this case drug discovery) can be reframed around data and data structures in partnership with industry leaders. domain.
Wild DJ. Mining large heterogeneous datasets in drug discovery. Expert Opinion on Drug Discovery 4(10):995–1004.
Wild DJ, Ding Y, Sheth AP, Harland L, Gifford EM, Lajiness MS. Systems chemical biology and the Semantic Web: what they mean for the future of drug discovery research. Drug Discovery Today 17:469–474.
Before anything can be reasoned over it has to exist, be accessible, and be AI-ready. This includes both technical data and human data. The FRST Challenge — an $8m NIST-funded prize competition run through the Crisis Technologies Innovation Lab — existed to generate precise in-building location data for first responders, a stream that simply did not exist before. The paper below (submitted) is about the collection of human data in the hight stress context of an incident response.
Rolfson A, Rosen A, Jennings C, Wild DJ. Using semi-structured interviews to elicit information needs and flows for pre-incident planning and technology implementation in emergency management and response. Journal of Homeland Security and Emergency Management, submitted.
Integrating heterogeneous sources, often messy data tables, into a single object that permits inference. Ontological annotation, entity resolution, and the representational choices that determine what can subsequently be computed.
Yang JJ, Gessner CR, Duerksen JL, Biber D, Binder JL, Ozturk M, Foote B, McEntire R, Stirling K, Ding Y, Wild DJ. Knowledge graph analytics platform with LINCS and IDG for Parkinson's disease target illumination. BMC Bioinformatics 23:37. Evidence-path analytics over an integrated knowledge graph, deliberately without model training, so that provenance stays inspectable where ground truth is uncertain or incomplete.
Passi A, Rajput NK, Wild DJ, Bhardwaj A. RepTB: a gene ontology based drug repurposing approach for tuberculosis. Journal of Cheminformatics 10:24. An ontology-enriched network of 26,404 edges over 6,630 drug and 4,083 target nodes, analysed by network-based inference.
Chen B, Dong X, Jiao D, Wang H, Zhu Q, Ding Y, Wild DJ. Chem2Bio2RDF: a semantic framework for linking and data mining chemogenomic and systems chemical biology data. BMC Bioinformatics 11:255.
Estimating what is absent, and treating the pattern of absence as evidence in its own right. Calibration matters here: a prediction is only usable if its uncertainty is legible to whoever must act on it.
Yang JJ, Goff A, Wild DJ, Ding Y, Annis A, Kerber R, Foote B, Passi A, Duerksen JL, London S, Puhl AC, Lane TR, Braunstein M, Waddell SJ, Ekins S. Computational drug repositioning identifies niclosamide and tribromsalan as inhibitors of Mycobacterium tuberculosis and Mycobacterium abscessus. Tuberculosis 146:102500. Semantic knowledge graph association combined with machine learning, carried through to wet-lab confirmed inhibitors with chemistry unlike existing TB drugs.
Seal A, Wild DJ. NetPredictor: R and Shiny package to perform drug–target network analysis and prediction of missing links. BMC Bioinformatics 19:265. General-purpose missing-link prediction for unipartite and bipartite networks; the method is stated independently of the drug discovery application.
Seal A, Ahn Y-Y, Wild DJ. Optimizing drug–target interaction prediction based on random walk on heterogeneous networks. Journal of Cheminformatics 7:40.
Chen B, Ding Y, Wild DJ. Assessing drug target association using semantic linked data. PLoS Computational Biology 8(7):e1002574. A semantic linked network of ~290,000 nodes and ~720,000 edges, and the Semantic Link Association Prediction (SLAP) algorithm for recovering unmeasured compound–target links.
Learned representations of structure, and the AI-adjacent methods that surround them — natural language processing, entity extraction, embeddings, and more recently large language models used both to query large graphs and to summarise for people working under time pressure. Current work on LLM-based and agentic systems is not yet published.
Gao Z, Fu G, Ouyang C, Tsutsui S, Liu X, Yang JJ, Gessner C, Foote B, Wild DJ, Ding Y, Yu Q. edge2vec: representation learning using edge semantics for biomedical knowledge discovery. BMC Bioinformatics 20:306. Embedding that treats edge type as carrying meaning, rather than embedding nodes in a graph whose relations are assumed homogeneous. Used as a descriptor source, it lifts downstream classifier performance substantially.
Features that persist across scales and samplings, relating to the underlying structures of the data.
Kulkarni VS, Wild DJ. An activity canyon characterization of the pharmacological topography. Journal of Cheminformatics 8:41. Treating an activity landscape as terrain, and reading its features — ridges, canyons — as the object of study.
Getting structure in front of an analyst, responder or commander in a form they can act on, and understanding when such systems are and are not trusted.
Obi I, Paul LJ, Liao W, Loukil M, Hayashi S, Comer M, Rogers CO, Wild DJ, Shih PC. Project APRED: a web-based data analytics platform for supporting community disaster resilience. Journal of Emergency Management 21(5):399–419. Computing jointly across risk, resilience and economic data for emergency and disaster planning.
Hill JJ, Wild DJ, Schmidt PM. Optimizing virtual health technologies through a performance-based readiness model: lessons from the field deployment of the National Emergency Tele-Critical Care Network. Military Medicine 188(Suppl 6):377–384.
Full list at Google Scholar.
Affiliations
DIRECTOR
Artificial intelligence for command, control and communications, and critical infrastructure technologies for first responders, cybersecurity and defense. Spanning basic research through to operational application. Successor to the Crisis Technologies Innovation Lab.
DIRECTOR
The methodological core: graph and semantic representation, topological structure, and inference under partial observation
Core faculty
Center of foundational and applied research in AI at Indiana University
Faculty fellow
Exploring applications in defense including international disaster response, safety, military medicine and situational awareness.
Faculty Fellow
Alongside faculty appointments in cybersecurity and data science at the Luddy School — the institutional home for the security applications of this work.
Lead Instructor
I am lead instructor or co-instructor for several applied project-based learning classes, including IU's Hacking for Defense program.
Contact