AI and oncology development
AI-Ready Cancer Research Starts With Data Another Team Can Verify
How connected cancer datasets, reproducible findings and reliable trial records support responsible AI research—and which benefits still need evidence.
By Lampgen Clinical Research

5 min read · estimated
What the source reported
AI-ready cancer research depends on usable, well-documented data, reproducible scientific findings and evaluation in the intended setting. Shared infrastructure can make cancer datasets easier to discover and analyze; it does not, by itself, establish a treatment benefit or make a model appropriate for clinical decisions. This distinction connects a planned SXSW discussion to practical decisions facing oncology research teams. [1–3]
SXSW’s October 27, 2026 program announcement includes “Can AI Cure Cancer? It Depends on the Data” among its 2027 sessions. The participant-written description supplied for this analysis discusses data silos, reproducibility and collaboration across academic, commercial and nonprofit organizations. This is a conference agenda, rather than a report of clinical results. Its optimism about future cures should be read as the participants’ framing, not a demonstrated outcome or an SXSW endorsement. [1]
What makes research data usable for AI?
NCI’s 2023 data-readiness request described characteristics including completeness, standards, computable formats, documentation, annotations, balance, privacy and security. Those remain useful questions for reviewing a proposed dataset. The notice was a time-limited consultation, not a current mandate or a certification that a dataset is ready for every AI application. [2]
Lampgen’s interpretation is that readiness is specific to the intended question. A dataset suitable for exploring a biological association may not support a model that recommends a treatment or identifies trial eligibility. Teams should define the intended use, necessary variables and acceptable missingness before choosing a model.
Shared infrastructure helps researchers connect evidence
NCI’s Cancer Research Data Commons connects cancer research datasets with analysis resources and common infrastructure. Its data-use policies also distinguish general access from controlled data and require uses consistent with applicable agreements and participant consent. Sharing therefore includes governance as well as technical availability. [3]
Connecting imaging, molecular and clinical records could help investigators examine relationships that are difficult to study in isolation. NCI describes this opportunity in multi-omics research while emphasizing expert review and validation. Potentially useful relationships still require biological interpretation and testing. [4]
Reproducibility remains a scientific test
In the 2021 Reproducibility Project: Cancer Biology analysis, researchers repeated 50 experiments from 23 selected high-impact papers and examined 158 effects. Results varied across assessment methods, with replication effects generally weaker than the original findings. The selected preclinical experiments do not provide an estimate of how much of all cancer science is reliable. [5]
The practical lesson is to retain the information needed to examine a result: methods, data versions, analytical code where sharing is permitted, and experimental conditions. Rerunning an analysis on the same data tests computational reproducibility. Repeating a biological finding with new experiments asks a different question. Neither alone establishes patient benefit.
Representation must be examined, even after identifiers are removed
NCI’s July 2026 discussion of generative AI warns that biased or incomplete population representation can affect research conclusions. Removing personal identifiers does not remove those evidence gaps. NCI describes reproducible workflows, evaluation for bias and review by domain experts as part of responsible research use. [4]
Our proposed review would compare performance across relevant patient groups, tumor subtypes, institutions and collection conditions, with uncertainty reported where samples are small. Race and ethnicity require careful interpretation alongside clinical characteristics and structural factors; they should not be treated as interchangeable biological explanations.
The possible impact is better research decisions
With stronger evidence, teams may be better able to prioritize hypotheses, select informative experiments and identify where another study is necessary. That could reduce avoidable rework. It could also expose previously hidden limitations and add necessary verification work. The net effect on development time and cost needs a defined comparison.
Benefits for immunotherapy, therapeutic cancer vaccines or other treatments depend on the specific discovery, its experimental support and subsequent clinical evaluation. A data platform or conference session cannot establish those benefits. Improving access also requires workable referral, enrollment and follow-up pathways.
What this means for trial teams
Preserve specimen context. Record the protocol-relevant collection time, treatment status, processing conditions and assay version. An illustrative problem is a response analysis that mixes pretreatment and post-progression samples without distinguishing them. Teams need scientific review of that difference, with documentation gaps traced to their source.
Make definitions and discrepancies visible. Align the meaning of key outcomes across the relevant records and vendors. Investigate missing or conflicting information and document the resolution. These are Lampgen’s proposed operational practices; they are not reported results from the SXSW panel.
Assign responsibilities at the handoff. CRAs can investigate site documentation gaps, escalate findings and verify follow-up. Study managers can coordinate recurring issues across sites and vendors. Data management, biostatistics, scientific leaders, medical monitors and investigators retain their respective decisions and responsibilities.
Evaluate the intended use before relying on an output. Ask which data supported the model, whether testing was independent of development, where performance is uncertain and what human review is required. For a trial decision, preserve the rationale and evidence that informed it. Reliable records support this evaluation; they do not guarantee an accurate model.
Lampgen Clinical Research’s oncology staffing perspective is that evidence quality is supported by specific work with clear ownership. Teams can review the related AURORA analysis, practical monitoring insight and oncology CRA service information below. This article does not imply an affiliation with SXSW or the organizations discussed.
Analysis by Tommy Lampley, Jr., M.S., Founder, Lampgen Clinical Research. Published November 13, 2026. The accompanying visual is an AI-generated conceptual editorial illustration. Its tissue, scan and record fragments are illustrative and contain no actual patient or study data.
Sources
- 1. SXSW, announced PanelPicker sessions for SXSW 2027, October 27, 2026
- 2. NCI, AI use cases and cancer data-readiness consultation, NOT-CA-24-002, October 27, 2023
- 3. NCI Cancer Research Data Commons, infrastructure and data-use policies
- 4. NCI, Questions Answered: Generative AI in Bioinformatics, July 8, 2026
- 5. Errington et al., Investigating the replicability of preclinical cancer biology, eLife, 2021
- Related Lampgen insight: AURORA AI co-scientists and cancer experiments
- Related Lampgen insight: evaluating AI alerts in oncology monitoring
- Lampgen oncology CRA staffing