A typical collaboration entails conducting a pilot project using banked, clinically annotated sample sets to investigate a clinical question of interest. Processing and analysis can usually be completed within a few weeks. If the initial analysis identifies a potential novel test, we continue with further validation in larger sample sets and prospective studies. At all stages of the project we work collaboratively, reporting and discussing our analyses and benefiting from the clinical expertise of our research partners.
Typical Course of Collaboration
Although Biodesix provides a flexible approach to research partnerships, collaborations typically follow a general pattern:
- Biodesix’ Business Development or Academic Collaborations Directors are available to discuss project feasibility and to frame the exact terms of the project to be undertaken. At this stage, sample requirements for pilot and validation studies as well as collaborative services are discussed.
- To discover a new classifier with the ability to separate samples into distinct clinical groups, we generally require 40-50 samples per group. The samples are divided into a training set, used for classifier development, and a test set, for classifier validation. It is desirable to have comprehensive clinical data to allow multivariate analysis and minimize bias in our analyses.
- MALDI mass spectra are obtained in our research laboratory from the samples provided. Biodesix can also work with MALDI spectra obtained by our Bio/Pharma partners or collaborators.
- Spectra are preprocessed using Biodesix’ proprietary ProTS software.
- Features in the processed mass spectra that can discriminate between the clinical groups are determined and evaluated for their statistical significance. This information can be used directly for biomarker studies.
- Features discriminating between the groups in the training set are used in the construction of a k-nearest neighbor classifier – a tool, which, when presented with a new spectrum, will attribute it to one of two groups. The quality of the classifier is tested directly on the training set using leave-one-out cross validation methods.
- The robustness of the classifier is assessed using the test set, not used in classifier development.
- Reports are drawn up on the details of the analysis, sent to our collaborators or Bio/Pharma partners, and discussed fully in a follow-up conference call or meeting.
- Any further analyses agreed upon during the course of the report review may be carried out and presented.
Further Validation and Prospective Studies
- If the pilot project is successful, the opportunities for further validation and prospective studies are discussed. If relevant, possible extensions of the original project can be undertaken.