We provide the example training set that was used to generate Figures 1a and 5.

There are three folders

• training_historic (16 runs from each of the historic products HP1 ... HP5)
• training_novel (4 runs from the novel product NP)
• testing (100 runs from the novel product NP)

Each folder contains a file experimental_conditions.csv that lists the process conditions. The subfolders HP1 ... HP5 or NP then contain experimental runs obtained with these conditions for the products.

For example training_historic/HP1/HP1_ExpHist00.csv contains the measurements obtained for the experimental condition ExpHist00 (which is found in training_historic/experimental_conditions.csv) with the product HP1. I.e. measurements for 6 quantities once per day. The file HP1_ExpHist00_details.csv further contains intermediate measurements throughout each day. This is NOT used for model training and included here to provide a better understanding of the bioprocess emulator.

• VCD (Viable Cell Density)
• Glc (Glucose Concentration)
• Gln (Glutamine Concentration)
• Amm (Ammonium Concentration)
• Lac (Lactate Concentration)
• product (concentration of the desired product, also known as titer)

• Xv_0, Glc_0, Gln_0, Amm_0, Lac_0 (initial concentrations)
• pH_0, T_0 (initial pH and Temperature)
• pH_switch_time, T_switch_time (time at which pH and Temperature switch)
• pH_switch, T_switch (Temperature and pH after switch)
• Glc_feed, Gln_feed (mass of Glucose and Glutamine in the feed)
• feed_start, feed_end (day at which the feeding starts and ends)
• feed_volume (volume of the feed)
• measurement_volume (once per day this amount of volume is removed from the bioreactor, simulates a measurement)
• stirring_rate
• initial_volume (initial reactor volume, changes through feeds and measurements)
• total_days (duration of the experiment)