Technology

Download PDF

Biovista’s B2B Services are all backed by a systematic discovery technology platform that is both extremely powerful and flexible.

This proprietary platform, called COSS™ (Clinical Outcomes Search Space), supports Biovista scientists in uncovering non-obvious correlations between drugs, molecular targets, pathways, adverse events and diseases and constructing evidence-based biological plausibility rationale on a systematic and highly predictable basis.

Biovista’s COSS™ platform is a hybrid approach, combining literature-based discovery with in silico simulations and resource mining to develop ranked lists of outcomes that answer a host of drug development questions.

At its core COSS™ creates multi-dimensional profiles of biologically relevant entities such as genes, pathways, diseases and adverse events from public and other sources. Biases are minimized through an include-all-information extraction process and the ability to analyze a problem from multiple viewpoints.

COSS™ adopts a ‘wisdom of the crowds’ stance, where every reference represents a vote of confidence of some, assumedly, reasonably savvy member of the crowd on the topic of interest and where the co-mentioning of two biologically relevant terms (such as a gene and a pathway, or a drug and a target) is taken as an indication of “a potentially interesting, to-be-confirmed correlation in some biological context”.

COSS™ multi-dimensional profiles are optimized for the type of problem being addressed and subsequently compared against each other. For example, in a typical repositioning scenario, the profile of a compound of interest is compared against the profiles of all possible indications, the degree of overlap of these profiles determining the relevance (ranking) of that compound for each disease.

In the context of discovery in a field where a lot of gaps exist in our body of knowledge, where there is a lot of controversy, me-too research, approximate predictive tools and where the system being study is characterized by ill-understood complexity, the COSS™ hybrid approach has the best chances of success because it integrates the outcomes of other non-deterministic tools and methodologies, thereby providing Mechanism-of-Action based overviews that can point experts ‘versed in the art of drug development’ in highly promising regions of what is a vast problem search space.

 

COSS™ components
  1. Natural Language Processing (to extract information from unstructured data sources)
  2. Ontologies (to organize information in semantically useful categories)
  3. Visualization (to help understand complex data sets)
  4. Data and Text Mining (to identify non-obvious information from massive data sources)
  5. Molecular docking and cheminformatics simulations to assess predicted outcomes

 

COSS™ is deployed over a growing internal database of scientific articles from major journals, patents, adverse events databases and other resources. With over 20 million records and billions of correlations on tap, it is currently one of the largest database of its kind in the world. COSS™ can extract actionable information in runs that take from seconds to a few minutes allowing subject matter experts to explore multiple ‘what if’ scenarios in a very short period of time.

Biovista’s technology components are combined into specific workflows, to provide turnkey solutions that maximize the value of research and clinical development.

 

COSS™ Validation

Biovista has done extensive studies on the predictive accuracy of COSS™ which have shown a Receiver Operating Characteristic (ROC) value of 0,75. This is well within the bounds for a commercially viable technology with Biovista’s own drug pipeline work supporting these findings.

 ROC values of COSS

 Comparative ROC values of diagnostic tests used in the clinic

 

The high ROC value of COSS™ allows Biovista’s subject matter experts to construct biological plausibility hypotheses with high Confidence in Rationale (CiR).

For more information

To find out more about the use of COSS™ in collaborative projects, you may visit the Research and Publication pages of this site.

Menu