Centralized scientific communities are less likely to generate replicable results

Area of Science:

• Biomedical research
• Toxicogenomics
• High-throughput screening

Background:

• Concerns exist regarding the reproducibility of scientific findings across various research fields.
• Large-scale evaluations of experimental robustness are lacking.
• Drug-gene interactions are crucial for understanding biological mechanisms and potential therapeutic targets.

Purpose of the Study:

• To evaluate the robustness of drug-gene interaction claims using large-scale data.
• To compare the replicability of findings reported in single versus multiple publications.
• To investigate the impact of scientific community structure on claim reproducibility.

Main Methods:

• Utilized drug-gene interaction claims from the Comparative Toxicogenomics Database.
• Linked claims with high-throughput experimental data from the LINCS L1000 program.
• Analyzed a subsample of interactions with multiple published findings to compare community structures.

Main Results:

• Claims reported in multiple papers showed significantly higher replication rates (45.5%) than expected.
• Claims reported in single papers replicated 19.0% more frequently than expected.
• Centralized scientific communities propagated less replicable claims than decentralized communities.

Conclusions:

• Replication across multiple studies enhances the robustness of scientific findings.
• Decentralized research communities with diverse methods and independent teams yield more reliable results.
• Policies promoting decentralized collaboration can improve the overall robustness of biomedical research.