Machine learning-assisted exploration of multidrug-drug administration regimens for organoid arrays
Abstract
Combination therapies enhance the therapeutic effect of cancer treatment; however, identifying effective interdependent doses, durations, and sequences of multidrug administration regimens is a time- and labor-intensive task. Here, we integrated machine-learning, automation, and large microfluidic arrays of cancer spheroids or patient-derived organoids formed in a tissue-mimetic hydrogel to achieve drastic acceleration of the discovery of effective multidrug administration regimens. For the clinically approved drug combination, we discovered a sequential administration regimen leading to a substantial reduction in the total drug dose, in comparison with concurrent drug supply, both at comparable drug efficacy. For the drugs that are currently under clinical development, we found a synergistic effect of concurrently administered drugs and showed that the synergy diminishes for the sequential drug supply. The developed strategy holds promise for the discovery of effective combination therapies for advanced cancer treatment, including personalized chemotherapies.
https://doi.org/10.5061/dryad.0vt4b8h8x
Overview
This dataset supports the study of optimizing multidrug administration regimens for organoid arrays using machine learning (ML)-guided strategies. The data include processed fluorescence imaging results representing cell viability in response to different drug combinations, administration sequences, and timing strategies.
The primary goal of the study is to identify optimal drug sequences that improve therapeutic efficacy using organoid-based models and Bayesian optimization (BO).
Description of the data and file structure
The data was collected from the fluorescence images of spheroids and organoids in the microarrays.
Data collection and processing
Fluorescence images of spheroids and organoids cultured in microfluidic arrays were collected to quantify live/dead cells. These images were processed using:
- ImageJ (NIH, USA)
- Custom Python software: Live-dead tool (DOI:10.5281/zenodo.14903369)
Optimization of drug regimens was performed using:
- Gryffin optimization software and experiment tracking:
ml-mf_chemo repository
(DOI:10.5281/zenodo.14890340)
Data access and additional sources
- This dataset is hosted on Dryad: https://doi.org/10.5061/dryad.0vt4b8h8x
- Additional files and code:
File Structure and Contents
Each folder corresponds to figures in the publication and contains experimental data used for machine learning-assisted optimization and validation of drug administration regimens.
/Fig 3 – Sequential Drug Administration to MCF-7 Spheroids
Fig 3b – Variation in Cell Viability Across Experimental Generations
gen– Generation/experimental series IDsample– Sample ID within generationcv_mean– Mean cell viability (%) from ≥3 independent experimentscv_sd– Standard deviation of cell viability (%)seq– Drug sequence identifier (e.g., a = 1-2-3, b = 1-3-2, etc.)
Fig 3c – Cell Viability for Tested Sequential Drug Schedules
gen– Generation/experimental series IDsample– Sample ID within generationcv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
Fig 3d – Validation of Best-Performing Drug Combinations
name– Sample label:G3_7: Simultaneous administration of sample 7 from generation 3GS2_7: Sequential administration of sample 7 from generation 2Ctrl: Control samples (no treatment)
cv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
/Fig 4 – Comparison of Sequential vs Individual Drug Administration
Fig 4a – Comparison of Simultaneous Multidrug vs Single-Drug Treatment
name– Sample label:G3_7: Concurrent 3-drug administrationDOX_G3_7: 1.25 µM DOX for 48hCPA_G3_7: 94 µM CPA for 48h5FU_G3_7: 105 µM 5-FU for 48hCtrl: Control samples (no treatment)
cv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
Fig S4b – Comparison of Sequential Multidrug vs Single-Drug Treatment
name– Sample label:GS2_7: Sequential 3-drug administrationDOX_GS2_7: 1.25 µM DOX for 8hCPA_GS2_7: 94 µM CPA for 8h5FU_GS2_7: 105 µM 5-FU for 32hCtrl: Control sample (no treatment)
cv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
/Fig 6 – Sequential Drug Administration to PDOs
Fig 6b – Variation in Cell Viability Across Sequential Schedules
gen– Generation/experimental series IDsample– Sample IDcv_mean– Mean cell viability (%) from ≥3 experimentscv_sd– Standard deviation of cell viabilityseq– Sequence of 2-drug administration (e.g., a = 1-2)
Fig 6c – Cell Viability in Tested Sequential Drug Schedules
gen– Generation/experimental series IDsample– Sample IDcv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
Fig 6d – Validation of Best Simultaneous vs Sequential Drug Combinations
name– Sample label:G3_2: Best-performing simultaneous combinationGS2_2: Best-performing sequential combinationCtrl: Control
cv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
/Fig 7 – Validation of Drug Administration on PDOs
Fig 7a – Simultaneous Drug Treatment vs Individual Drugs
name– Sample label:G3_2: Best simultaneous combinationOLA_G3_2: 116 µM Olaparib for 96hIBET_G3_2: 0.993 µM IBET-762 for 96hCtrl: Control
cv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
Fig 7b – Sequential Drug Treatment vs Individual Drugs
name– Sample label:GS2_2: Best sequential 2-drug treatmentOLA_GS2_2: 116 µM OLA for 30hIBET_GS2_2: 0.993 µM IBET-762 for 66hCtrl: Control
cv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)
/Supporting Data/
Fig S4a-b – Reference Cell Viability Graph & Protocol for Drug Efficacy Assessment
cv– Cell viability in % (0–100%)green– Calcein fluorescence intensity (arbitrary units, a.u.), GFP channel- At 0% viability: background signal
- At 100% viability: maximum signal
red– Propidium Iodide fluorescence intensity (a.u.), GFP channel- At 0% viability: maximum signal
- At 100% viability: background
cells– Cell type:MCF-7– human breast cancer cell line (panel A)dcbxto.58– patient-derived breast cancer organoid (PDO) line (panel B)
Fig S5a – Dose-Response Curves for MCF-7 Spheroids to Individual Drugs
drug– Drug label:DOX= Doxorubicin5-FU= 5-FluorouracilCPA= Cyclophosphamide
conc– Drug concentration (µM)cv_mean– Mean cell viability (%) from ≥3 independent experimentscv_sd– Standard deviation of viability (%)
Note: The 10000 µM 5-FU sample was each tested in a single independent experiment; therefore, standard deviations are not available (NA). The data point was included solely to aid visualization of the dose–response trend and was not used in curve fitting.
Fig S5b – Comparison of Live/Dead Assay and Ki-67 Proliferation Assay for DOX
drug– Drug label (same as above)conc– Drug concentration (µM)feas– Feasibility of preparation (TrueorFalse)cv_exp– Experimental cell viability (%) from a single experiment (mean value of up to 100 spheroids)mode– Assay type:ld= Live/Dead assayki67= Proliferation marker using FITC-conjugated Ki-67 antibody
Fig S7a-c – Concurrent Drug Administration to MCF-7 Spheroids
gen– Generation/experimental series IDsample– Sample ID within generationconc_dox– DOX concentration (µM)conc_cpa– CPA concentration (µM)conc_5-fu– 5-FU concentration (µM)cv_mean– Mean cell viability (%) from ≥3 independent experimentscv_sd– Standard deviation of cell viabilityci_mean– Combination Index (CI) mean from ≥3 independent experimentsci_sd– Standard deviation of CI
Fig S9b – Total Drug Dose in Sequential vs Simultaneous Schedules (MCF-7)
gen– Generation/experimental series IDsample– Sample IDseq– Drug sequence (e.g., a = 1-2-3)dox_dose– DOX dose (%) relative to simultaneous schedulecpa_dose– CPA dose (%)5-fu_dose– 5-FU dose (%)dose– Total relative dose (%) of 3-drug combination compared to simultaneous administration
Fig S13 – Dose-Response for PDOs Treated with OLA or IBET-762
drug– Drug name:OLA(Olaparib),IBET-762Conc– Drug concentration (µM)cv_mean– Mean cell viability (%) from ≥3 independent experimentscv_sd– Standard deviation of cell viability
Note: The 100 µM and 2000 µM IBET-762 samples were each tested in a single independent experiment; therefore, standard deviations are not available (NA). These data points were included solely to aid visualization of the dose–response trend and were not used in curve fitting.
Fig S14a-c – Concurrent Drug Administration to PDOs
gen– Generation/experimental series IDsample– Sample IDconc_ola– OLA concentration (µM)conc_ibet762– IBET-762 concentration (µM)cv_mean– Mean cell viability (%) from ≥3 independent experimentscv_sd– Standard deviation of cell viabilityci_mean– Combination index (CI) mean from ≥3 independent experimentsci_sd– Standard deviation of CI
Fig S15a-b – Sequential Drug Schedules for OLA/IBET-762 (PDOs)
gen– Generation/experimental series IDsample– Sample IDseq– Sequence of 2-drug administration (e.g., a = 1-2)ola_time– Duration of OLA administration (hours)ibet_time– Duration of IBET-762 administration (hours)dose– Total dose (%) compared to simultaneous administration
Variable Definitions
| Variable | Description |
|---|---|
gen |
Generation/experimental series ID |
sample |
Sample ID within each generation or experimental batch |
cv_mean |
Mean cell viability (%) from at least 3 independent experiments |
cv_sd |
Standard deviation of cell viability (%) across experiments |
cv_exp |
Experimental viability (%) from a single experiment (mean of up to 100 spheroids) |
seq |
Drug sequence identifier (e.g., a = 1-2-3, b = 1-3-2, etc.) |
name |
Label for experimental condition (e.g., G3_7, GS2_7), includes drug combination and admin strategy |
green |
Calcein fluorescence intensity (a.u.) – indicates live cells (GFP channel) |
red |
Propidium Iodide fluorescence intensity (a.u.) – indicates dead cells (GFP channel) |
cells |
Cell type used in the assay (e.g., MCF-7 breast cancer or dcbxto.58 breast PDOs) |
drug |
Drug name (e.g., DOX = Doxorubicin, 5-FU = 5-Fluorouracil, CPA = Cyclophosphamide, OLA, IBET-762) |
conc |
Drug concentration in micromolar (µM) |
feas |
Feasibility of dilution – True if feasible, False if not |
mode |
Assay mode: ld = live/dead staining, ki67 = Ki-67 proliferation marker |
conc_dox |
Doxorubicin concentration in µM |
conc_cpa |
Cyclophosphamide concentration in µM |
conc_5-fu |
5-Fluorouracil concentration in µM |
conc_ola |
Olaparib concentration in µM |
conc_ibet762 |
IBET-762 concentration in µM |
ci_mean |
Mean combination index from a single experiment |
dox_dose |
Relative dose of Doxorubicin used in the schedule (%) |
cpa_dose |
Relative dose of Cyclophosphamide used (%) |
5-fu_dose |
Relative dose of 5-FU used (%) |
dose |
Total relative drug dose (%) compared to simultaneous administration |
ola_time |
Duration of Olaparib administration in hours |
ibet_time |
Duration of IBET-762 administration in hours |
Abbreviations and Notes
- ML: Machine Learning
- BO: Bayesian Optimization
- PDO: Patient-Derived Organoid
- MCF-7: Human breast cancer cell line
- DOX: Doxorubicin
- CPA: Cyclophosphamide
- 5-FU: 5-Fluorouracil
- OLA: Olaparib
- IBET-762: BET bromodomain inhibitor
- CI: Combination Index (CI < -0.1: synergy, CI = 0: additive effect, CI > 0.1: antagonism)
- a.u.: Arbitrary Units
- G3_X / GS2_X: Sample X from Generation 3 (G = simultaneous, GS = sequential schedule)
- Live/Dead Assay: Fluorescent viability assay using Calcein-AM (live, green) and Propidium Iodide (dead, red) Live-dead tool (DOI:10.5281/zenodo.14903369)
Access information
Other publicly accessible locations of the data: