cs.SEMar 13, 2026

An Empirical Investigation of Pre-Trained Deep Learning Model Reuse in the Scientific Process

Authors: Nicholas M. Synovic, Karolina Ryzka, Alessandra V. Vellucci Solari, Kenny Lyons, James C. Davis, George K. Thiruvathukal

Organizations: Computer Science Department, Loyola University Chicago, Chicago, IL, USA · Electrical and Computer Engineering Department, Purdue University, West Lafayette, IN, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Deep learning has achieved recognition for its impact within natural sciences, yet the prohibitive financial and technical cost of training models from scratch inhibit adoption. Following software engineering community guidance, natural scientists are reusing pre-trained deep learning models (PTMs) to amortize these costs. While prior works recommend PTM reuse patterns, we present the first empirical study of PTM reuse patterns in the natural sciences, quantifying the utilization and impact of PTM reuse within the scientific process across 17,718 peer reviewed, open access papers. Our results show that "Biochemistry, Genetics and Molecular Biology" has outpaced other natural scientific fields in PTM reuse, "adaptation" reuse is the most prevalent PTM reuse pattern identified across all natural science fields, and the "testing" stage of the scientific process has been most impacted by PTM integration.

Figures & tables

Fig. 1: PTMs support every stage of the scientific process — from observation and hypothesis generation through testing and evaluation. A supply chain of frameworks, fine tuning algorithms, and model registries enables scientists to apply data driven computational methods in their research.
Reuse PatternDefinitionExamples
ConceptualReplicate or reengineer PTMs described in academic literature.[ 19 , 20 , 21 ]
AdaptationModify existing PTMs to address different tasks.[ 22 , 23 , 24 ]
DeploymentEngineer existing PTMs for use in different computational frameworks.[ 25 , 26 , 27 ]
TABLE I: PTM reuse patterns identified in our study, with definitions and representative papers leveraging each pattern.
Fig. 2: Our methodology for identifying and analyzing peer reviewed, open access natural-science publications in megajournals. Publications retrieved from megajournal databases are enriched with OpenAlex metadata and analyzed for DL use, PTM reuse, and PTM reuse patterns. Manually curated ground truth labels guide refinement of the prompting strategy.
Fig. 3: Results of the mega-journal paper filtering process. Our queries returned 17,718 total unique papers between 2000 and 2025, of which 17,364 where indexed by OpenAlex. Of these, 14,998 had citations and 7,212 were natural science papers. 6,956 of these were available in JATS XML format.
OpenAlex Topic FieldExamplesProp.
Biochemistry, Genetics and Molecular Biology[ 61 , 62 , 63 ]36.91%
Environmental Science[ 64 , 65 , 66 ]19.31%
Neuroscience[ 67 , 68 , 69 ]18.59%
Agricultural and Biological Sciences[ 70 , 71 , 72 ]12.18%
Earth and Planetary Sciences[ 73 , 74 , 75 ]4.09%
Immunology and Microbiology[ 76 , 77 , 78 ]3.48%
TABLE II: The eight identified natural science OpenAlex topic fields, their proportions in our study population, and example papers for each topic.
Figure 6
Figure 7
Fig. 8: Counts of PTM reuse patterns per year, 2012–2025. Adaptation reuse is the most common pattern and grows steadily over time, while deployment reuse appears less frequently and conceptual reuse remains relatively rare.
Fig. 9: Scientific process stages affected by PTM reuse from 2012 to 2025. Annual counts show PTM reuse across five stages of the scientific process: observation, background, hypothesis, analysis, and testing. Reuse occurs most frequently during testing, while the other stages show comparatively fewer instances.
Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Paper TitleOpenAlex Topic 1OpenAlex Topic 2OpenAlex Topic 3Filtered
Characterization of RNA polymerase II trigger loop mutations using molecular dynamics simulations and machine learning [ 21 ]Biochemistry, Genetics and Molecular BiologyBiochemistry, Genetics and Molecular BiologyBiochemistry, Genetics and Molecular BiologyTrue
The genome of the migratory nematode, Radopholus similis , reveals signatures of close association to the sedentary cyst nematodes [ 27 ]Agricultural and Biological SciencesAgricultural and Biological SciencesAgricultural and Biological SciencesTrue
Generalizing biological surround suppression based on center surround similarity via deep neural network models [ 20 ]NeuroscienceNeuroscienceNeuroscienceTrue
Integration of smart sensors and phytoremediation for real-time pollution monitoring and ecological restoration in agricultural waste management [ 66 ]Environmental ScienceEnvironmental ScienceEnvironmental ScienceTrue
A semi-supervised Bayesian approach for simultaneous protein sub-cellular localisation assignment and novelty detection [ 98 ]ChemistryChemistryBiochemistry, Genetics and Molecular BiologyTrue
Progress and prospects for research on Martian topographic features and typical landform identification [ 99 ]Physics and AstronomyPhysics and AstronomyPhysics and AstronomyTrue
Appendix
TABLE III: Example papers from our initial search query with at least one citation filtered on the natural science topics they were assigned by OpenAlex.
More in this topic

AI-Assisted Scientific Research

AI-assisted scientific research uses machine learning and AI agents to support discovery, analysis, experimentation, and scientific communication.

See all 253 papers

Explore similar work

CardsList

  1. Measuring research data reuse in scholarly publications using generative artificial intelligence: Open Science Indicator development and preliminary results

    Apr 30, 2026Lauren Cadwallader, Iain Hrynaszkiewicz, parth sarin +1Large Language Model-Based Text Classification

    PLOS, Cambridge, UK · DataSeer, Vancouver, Canada

  2. The Shrinking Lifespan of LLMs in Science

    Apr 8, 2026Ana TrišovićAI-Assisted Scientific Research

    Computer Science & Artificial Intelligence Laboratory, Massachusetts Institute of Technology, 32 Vassar St, Cambridge, MA 02139, USA

  3. Persistence Paradox in Dynamic Science: Evidence from the Deep Learning Revolution

    Jun 28, 2025Honglin Bao, Beichen Lu, Kai Li

    Data Science Institute, University of Chicago · School of Information Sciences, University of Tennessee, Knoxville