SOURCE DATA: FIVE-INPUT INFERENCE FROM THREE TIME-RESOLVED READOUTS

Paper
Engineering microbial consortia for distributed signal processing
Nature Communications manuscript NCOMMS-26-017824B

Folder / archive
07_Supplementary_Figure_17

Dataset description
This archive contains the experimental plate-reader time courses used for Supplementary Figure 17. A two-strain Escherichia coli community generated three time-resolved readouts—optical density (OD), YFP, and mCherry—in response to five chemical inputs:

- tetrathionate (TTR; workbook column "ttr")
- isopropyl beta-D-thiogalactopyranoside (IPTG; workbook column "iptg")
- kanamycin (workbook column "kan")
- chloramphenicol (workbook column "cm")
- glucose (workbook column "glucose")

The community contained an IPTG-responsive mCherry strain and a TTR-responsive YFP strain. Kanamycin, chloramphenicol, and glucose were inferred from their reproducible effects on the shared growth and fluorescence trajectories rather than from dedicated reporter channels.

Across the five workbooks, the archive contains:
- 453 two-strain co-culture time-course samples
- 176 unique five-input concentration combinations
- 288 recorded time points per sample
- approximately 24 h of observation at approximately 5-min intervals

Replicate coverage varies across input combinations because the dataset combines multi-input combination plates, targeted single-input dose-response plates, and repeated no-input controls. Blank and water-control rows are retained in the workbooks but are not included in the reported total of 453 co-culture samples.

Files
1. 1029update_2025-09-24_5input3output_plate1.xlsx
   - 95 co-culture samples
   - 95 unique five-input combinations
   - 1 blank well
   - 288 time points spanning 0-86,109.6 s

2. 1029update_2025-09-24_5input3output_plate2.xlsx
   - 95 co-culture samples
   - 95 unique five-input combinations
   - 1 blank well
   - 288 time points spanning 0-86,108.1 s
   - contains an auxiliary worksheet named "fix"; this worksheet is not a measurement channel

3. 1029update_doseresponse_5input3output.xlsx
   - 84 co-culture samples
   - 36 unique conditions
   - targeted single-input dose-response measurements plus repeated no-input controls
   - 8 water-control wells and 4 blank wells
   - 288 time points spanning 0-86,110.7 s
   - contains an auxiliary worksheet named "template"; this worksheet is not a measurement channel

4. 2025-11-19_processed_plate1_doseresponse_bigbird.xlsx
   - 84 co-culture samples
   - 36 unique conditions
   - targeted single-input dose-response measurements plus repeated no-input controls
   - 4 blank wells
   - 288 time points spanning 0-86,108.3 s

5. 2025-11-19_processed_plate2_K.xlsx
   - 95 co-culture samples
   - 95 unique five-input combinations
   - 1 blank well
   - 288 time points spanning 0-86,108.7 s

Because many conditions are repeated across workbooks, the combined number of unique input combinations is 176 rather than the sum of the workbook-level counts.

Workbook organization
Every workbook contains the following three measurement worksheets:
- OD: optical-density time courses
- YFP: yellow-fluorescence time courses
- mCherry: red-fluorescence time courses

Within each workbook, these three worksheets have identical sample-row ordering and metadata. Each row corresponds to one microwell.

The first seven columns contain:
1. Sensor: sample type ("both" for the two-strain community; "blank" or "water" for controls)
2. ttr: TTR concentration
3. iptg: IPTG concentration
4. kan: kanamycin concentration
5. cm: chloramphenicol concentration
6. glucose: glucose concentration
7. Time [s]: microwell identifier

Columns 8 onward contain measurements at successive time points. The column headers are elapsed time in seconds.

The numerical concentration values are preserved exactly as recorded in the experimental workbooks. The workbook column names do not encode concentration units; units should be interpreted according to the experimental protocol and the associated analysis notebook.

Figure mapping
- Supplementary Figure 17a:
  Schematic of the two-strain, five-input system; no numerical source data are required for the schematic itself.

- Supplementary Figure 17b:
  VAE reconstruction analysis derived from the OD, YFP, and mCherry trajectories in all five workbooks.

- Supplementary Figure 17c:
  Representative reconstructed and experimental trajectories selected from the same dataset.

- Supplementary Figure 17d:
  Experimental training and validation predictions for TTR, IPTG, kanamycin, chloramphenicol, and glucose derived from the 453 co-culture time courses.

The VAE reconstructions, simulated trajectories, latent representations, and predicted concentrations are computationally derived from the experimental measurements and can be regenerated using the archived code, fitted parameters, trained models, and saved data partitions.

Controls and exclusions
Rows labeled "both" are the experimental co-culture samples counted in the reported dataset total. Rows labeled "blank" or "water" are retained as experimental controls but are excluded from the 453-sample total.

No rows in these five workbooks are explicitly labeled as contaminated, leaked, outliers, or otherwise excluded. Analysis-specific normalization, zero-value rescaling, training-validation partitioning, and model fitting are documented in the associated notebook.

Plate-reader calibration
Cross-instrument conversion tables used during preprocessing, where applicable, are supplied separately in the 08_Plate_reader_calibration Source Data archive.

Code and reproducibility
Analysis code is available at:
https://github.com/youlab/multiplexed_sensing

Relevant notebook:
supplementary_fig_17.ipynb

The analyses were performed using Python 3.12.7.

Contact
Lingchong You
Department of Biomedical Engineering, Duke University
Email: you@duke.edu
