CHEMOLYTICA

ML-Guided Catalyst Screening for Ester Hydrogenation

An undergraduate thesis, rewritten as a lab report: the data, the models, and what the errors say about the chemistry.

At a glance
Status
Completed
Context
Undergraduate thesis, chemistry
Question
Can machine learning predict the yield of an ester hydrogenation, and so help choose catalysts to test?
Data
[n] reactions from [source]
Methods
[Models compared]
Tools
Python · [libraries used]
Code and data
GitHub repository
Abstract

[A one-paragraph abstract: the question, the data, the method, the main result and what it means for choosing catalysts.]

1Background

Ester hydrogenation turns esters into alcohols, and the catalyst largely decides how well it works. Testing catalysts one by one is slow and expensive. If a model could estimate the yield before the reaction is run, it could help decide which catalysts are worth the bench time.

[A paragraph on what is already known, and the gap this project addresses.]

2Data

[A paragraph describing where the reactions came from, how many there were, which features describe each reaction, and how the data were cleaned.]

3Methods

The data were split into a training set and a held-out test set. Each model was fitted on the training set only, then judged on reactions it had never seen. Listing A.1 shows the core of that step.

Listing A.1 The core of the modelling step, simplified. The full pipeline is in the repository.
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_absolute_error, r2_score

df = pd.read_csv("reactions.csv")
X = df.drop(columns=["yield"])
y = df["yield"]

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=0)

model = RandomForestRegressor(random_state=0).fit(X_train, y_train)
pred = model.predict(X_test)
print(mean_absolute_error(y_test, pred), r2_score(y_test, pred))

[A paragraph on the models compared, how their settings were chosen, and how features were represented.]

4Results

[A paragraph stating the main result in plain words, with the key number.]

Experimental yield (%) Predicted yield (%)
Figure A.1

Predicted against experimental yield for the test set. Points on the dashed line are predicted exactly; the highlighted point is discussed in Section 5. Illustrative; replace with the real plot.

Table A.1 Test-set performance of each model.
ModelMAE (%)R2R^2
[Baseline][ ][ ]
[Model 2][ ][ ]
[Model 3][ ][ ]

5Discussion

[What the results mean for choosing catalysts. Discuss the highlighted point in Figure A.1: why the model missed it, and what that says about the chemistry.]

[What the results mean for choosing catalysts. Discuss the highlighted point in Figure A.1: why the model missed it, and what that says about the chemistry.]

Limitations

[Where the model should not be trusted: the size of the dataset, the range of catalysts covered, and any bias in the reactions that get reported.]

Reproduce this project
git clone [repository URL]
cd [repository]
pip install -r requirements.txt
jupyter nbconvert --to notebook --execute analysis.ipynb

References

  1. [Author, A. Title. Journal Volume, pages (Year).]

  2. [Author, B. Title. Journal Volume, pages (Year).]

Cite this project
Solomon, M. B. (2026). ML-Guided Catalyst Screening for Ester Hydrogenation. Chemolytica, PRJ 001.

Found an error or have a question about this project? Let me know.

REVISIONS8 November 2026 · First published.