KeyError 'iteration' in directed_evolution_simulation when running DMS workflow

[iRetamal]

Hi,

I'm trying to run the DMS workflow using evolvepro, but I encounter a KeyError: 'iteration' in the directed_evolution_simulation function.

Setup:

evolvepro environment activated (Python 3.11, Pandas 2.x)

Labels CSV file contains the columns:

variant,iteration,activity
E2A,0,-0.339226193
E2C,0,-0.393733896
E2D,0,-0.232532833
E2F,0,-0.181163852
E2G,0,-0.65668517
...

Error traceback (simplified):

Traceback (most recent call last):
  ...
  File "model.py", line 130, in top_layer
    iteration = labels_pd["iteration"]
KeyError: 'iteration'

Observation:

The iteration column exists but all values are 0.

Reading the CSV with pd.read_csv shows the correct column names.

Question:
Should the iteration column in labels have multiple iteration numbers already, or is it expected to be generated automatically by the workflow?

Thanks for your guidance!

[mat10d]

Can't really tell where that data you are showing is being passed in, but you want to follow a structure like in the colab/ folder example: https://github.com/mat10d/EvolvePro/blob/main/colab/rounds_data/kelsic_Round1.xlsx. Should have an excel file with Variant, activity, where the variant only has the position and the AA changed to. E.g. E2A should be 2A. Can you try reformatting to that and let me know how that goes?

[iRetamal]

Additional issue: undocumented label requirements & --measured_var not respected in DMS workflow

Hi again, I just got around to checking this again, apologies for the delay. Following up on this issue after getting past the embeddings/labels alignment problem.

Once the DMS workflow (scripts/dms/dms_main.py) starts running, it fails due to missing label columns that are not documented, which revealed a mismatch between the README and the actual implementation.

The README suggest:

--measured_var: Activity type to train on (options: activity, activity_scaled)

Based on this, my labels CSV only contains two columns (variant and activity), following the example in the previous answer. I also explicitly set --measured_var activity.

However, in evolvepro/src/model.py, inside top_layer(), the code always accesses the following columns:

y_train = labels[iteration.isin(iter_train)][measured_var]
y_train_activity_scaled = labels[iteration.isin(iter_train)]["activity_scaled"]
y_train_activity_binary = labels[iteration.isin(iter_train)]["activity_binary"]

In the practice: the regression model is always trained on y_train_activity_scaled, measured_var does not control the actual training target, and activity_binary is required even if the user is not running a binary task.

As a result, the workflow crashes unless the labels file contains all of the following columns, even when not documented or requested: activity, activity_scaled and activity_binary.

To get the DMS workflow running on my side, I ended up defining the additional label columns as follows (just to make the assumptions explicit):

• activity_scaled: z-score normalization of the raw activity values across the full dataset
• activity_binary: simple thresholding of activity (e.g. activity > 0 → active, otherwise inactive)

With these additions, the Step 3 DMS workflow start running without errors.

I wanted to check whether this preprocessing is consistent with the intended use of EVOLVEpro, or if there is a recommended way to define activity_scaled and activity_binary for DMS datasets. Clarifying this in the documentation could also help new users avoid confusion when preparing their labels.