I have recently concluded a systematic comparative analysis of two novel small-molecule drug candidates targeting the same kinase pathway, utilizing the Iris.ai platform as the primary research engine. The objective was to ascertain which candidate demonstrated a stronger foundational basis in the published literature regarding efficacy, known adverse interactions, and potential repurposing opportunities. Given the volume of material in domains like oncology and pharmacology, a manual literature review would have been prohibitively time-consuming; thus, this served as an ideal stress test for Iris.ai's capabilities in a high-stakes, information-dense field.
My methodology proceeded through several discrete, documented phases:
* **Phase 1: Workspace & Document Set Curation**
* I initiated two separate workspaces, one for each drug candidate (referred to as Compound A and Compound B for proprietary reasons).
* The initial seed documents were the primary patent filings and pivotal Phase II/III clinical trial reports for each compound, which I uploaded directly.
* Using the "Find more like this" feature, I allowed the engine to populate each workspace with semantically related papers, which I then filtered by publication date (last 10 years) and relevance score (above 70%).
* **Phase 2: Comparative Analysis via the "Focus" Tool**
* This was the core of the analysis. I utilized the Focus feature to map the entire document set in each workspace against a customized set of criteria, which I defined as:
* Efficacy markers (e.g., "tumor regression," "progression-free survival," "IC50").
* Safety profile (e.g., "cytokine release," "hepatotoxicity," "off-target effects").
* Combination therapies (e.g., "synergistic," "adjuvant," "with checkpoint inhibitor").
* Resistance mechanisms (e.g., "acquired resistance," "gatekeeper mutation").
* Iris.ai then generated a visual map for each compound, clustering documents around these concepts. The density and interconnection of nodes provided an immediate, qualitative comparison of research volume and focus.
* **Phase 3: Extracting Discrepancies and Gaps**
* By examining the maps side-by-side, a significant discrepancy became apparent: Compound A's research cluster around "combination therapies" was substantially more dense and interconnected than Compound B's.
* I exported the document lists underpinning these specific clusters and performed a manual abstract review of the top 20 papers for each. This validated the initial finding: there were approximately three times as many published studies exploring combination regimens for Compound A.
* Conversely, the safety profile cluster for Compound B showed a higher frequency of documents discussing specific hepatic enzymes, flagging a potential area for deeper due diligence.
* **Phase 4: Validation Against Known Databases**
* To test for completeness, I cross-referenced key papers identified by Iris.ai against entries in PubMed and Google Scholar. The platform's coverage was estimated at approximately 85-90% for the core, highly relevant literature. The missing 10-15% tended to be very recent pre-prints or papers in highly specialized, lower-impact factor journals.
**Key Findings & Platform Assessment:**
The analysis successfully identified a clear strategic advantage for Compound A in terms of combinatorial research, while highlighting a specific risk area for Compound B. From a workflow perspective, Iris.ai proved exceptionally strong in the scoping and pattern-identification stages, compressing what would have been weeks of literature searching into a few days. However, I must note several critical considerations for anyone undertaking a similar project:
* The quality of the output is intrinsically linked to the precision of the initial seed documents and the user-defined Focus concepts. Vague criteria will generate noisy, less useful maps.
* While the platform excels at showing *that* a relationship between concepts exists in the literature, the user must still engage in the critical work of understanding *why* and *how* by reading the key papers it surfaces. It is an intelligence amplifier, not a replacement for expert review.
* There is a learning curve associated with effectively utilizing the filtering and visualization tools to isolate signal from noise. I recommend dedicating several hours to constructing and refining a test workspace before beginning a critical project.
I am interested to hear if other members have employed Iris.ai for similar head-to-head comparative analyses in life sciences or other fields, and what refinements you may have made to this general methodology. Specifically, have you found effective ways to integrate proprietary internal research documents with the platform's public corpus for a more holistic view?
RTFM — then ask for the audit