Multi-Modal & Integrative Studies

Recent developments in Health Information Systems and high-dimensional molecular assays have led in recent years an exponential increase in the volume and complexity of routinely collected patient data as well of data collected in research studies involving human and animal models or low-level organism data. Simultaneously there has been an explosion of freely as well as commercially accessible databases, such as dataset repositories, pathway databases, sequenced and annotated genome and proteome databases, indexed bibliographies, etc. While traditional models of conducting research involve collection and analysis of one or a limited number of types of data, a new paradigm, that of multi-modal and integrative studies, is emerging calling for very sophisticated computational and data analytics and informatics required for effective design, execution and interpretation of such studies.

We indicatively describe types and examples of multi-modal and integrative studies. CHIBI faculty have substantial experience and methodological know how in these types of studies.

1. Clinical Trials incorporating clinical and high throughput assay molecular information . An indicative example related project at NYU involves personalized medicine molecular profiles of a clinical trial of colorectal cancer treatment that integrates proteomic and clinical information (PI. Dr. D. Cohen). Another example involves molecular profiles for determining responders in a clinical trial of patients with keloids (PI. Dr B. Cronstein).
2. Case-control and other non-randomized studies combining traditional clinical, epidemiological and high throughput genomic and proteomic data. An indicative example project executed by CHIBI faculty involves the determination of lung cancer subtype using gene expression microarray, and MALDI MS tumor and serum proteomic data (PI, Dr, P. Massion, at Vanderbilt University). In another example, CHIBI faculty analyzed combined SNP and epidemiological data to create signatures for the diagnosis of esophageal cancer and identified significant SNPs adjusting for the effect of recorded environmental factors. Currently at NYU Dr M. Blaser and his faculty are leading several studies where deep sequencing microbiomic and clinical  data are to be analyzed in tandem and then linked to diagnostic and outcome phenotypes.
3. Combining traditional Clinical Trial Data from several pre-existing trials to develop outcome prediction models from one trial and to validate against data from subsequent trials. An indicative example project executed by CHIBI faculty involves the development of several risk assessment models in patients with ARDS with clinical and proteomic data from one trial and validate against other trials (PI. Dr. L. Ware at Vanderbilt University).
4. Mapping high throughput data from one dataset to other datasets. Examples projects include several benchmarking and validation studies (PIs. Dr. Aliferis and Dr. Statnikov) in which molecular signatures for cancer diagnosis and outcome prediction were developed from gene expression microarray datasets and then they were validated in independently created datasets (from different labs, sometimes with different array platforms).
5. Combining evidence from experimental data with evidence from the literature, pathway databases, etc. An example project (PI, Dr E. Fisher) includes the determination of local pathways involved in progression and regression of atherosclerosis using a combination of denovo construction of pathways from gene expression microarray data, inducing all equivalence model classes that fit the data equally well and eliminating those that contradict prior experimentally validated pathways.

These are just a few of many examples of multi-modal data projects possibilities. CHIBI faculty are leading developers of new methods in this area and provide cutting-edge informatics support to researchers.

Contacts: Constantin Aliferis, Jinhua Wang, Yuval Kluger, Alexander Statnikov