This project implements and evaluates artificial neural networks (ANNs) to classify mice into eight experimental groups based on protein expression and modifications detected in the nuclear fraction of the cortex.

The classification task involves 8 classes determined by three factors:

• Genotype: control (c) or trisomic (t).
• Behavioral condition: context-shock (CS) or shock-context (SC).
• Treatment: memantine (m) or saline (s).

The 8 classes are:

1. c-CS-s: control – context-shock – saline
2. c-CS-m: control – context-shock – memantine
3. c-SC-s: control – shock-context – saline
4. c-SC-m: control – shock-context – memantine
5. t-CS-s: trisomic – context-shock – saline
6. t-CS-m: trisomic – context-shock – memantine
7. t-SC-s: trisomic – shock-context – saline
8. t-SC-m: trisomic – shock-context – memantine

The objective is the implementation and evaluation of fully connected neural networks (dense layers) for multiclass classification.

• data3.csv: Protein expression levels (1080 samples × 73 columns).
  • Column 1: sample identifier.
  • Columns 2–73: protein expression values.
• class3.csv: Class labels (1080 samples × 1 column, values 1–8).

No missing values are present in the dataset.

• Summary statistics were computed for all protein features.
• Class distribution was visualized (class_distribution.png).
• Protein correlation analysis was performed (correlation_matrix.png).

Outputs:

• summary_statistics.csv
• class_distribution.png
• correlation_matrix.png

• Removal of sample ID column.
• Normalization of protein features using Min-Max scaling to [0, 1].
• Train-test split: 67% training (723 samples), 33% test (357 samples).
• Encoding of class labels from 1–8 to integers 0–7.

Normalized data is saved in data_normalized.csv.

• Input layer: 72 protein features.
• Hidden layer: Dense(35) with ReLU activation.
• Dropout layer: Dropout(0.2).
• Output layer: Dense(8) with softmax activation.

Parameters:

• Trainable parameters: 2,843
• Optimizer: Adam
• Loss function: Sparse categorical crossentropy
• Epochs: 50
• Validation split: 20%

Outputs:

• learning_curve_model_1.png
• confusion_matrix_model_1.png

• Deeper fully connected neural network with additional dense layers.
• Same optimizer, loss function, and training procedure as Model 1.

Outputs:

• learning_curve_model_2.png
• confusion_matrix_model_2.png

Models were evaluated using:

• Accuracy on test set
• Confusion matrix
• Precision, recall, and F1-score

Visual outputs:

• Training vs validation learning curves
• Confusion matrices

• Python 3.9+
• pandas
• numpy
• matplotlib
• seaborn
• scikit-learn
• tensorflow

Install dependencies:

pip install -r requirements.txt

1. Clone the repository:

git clone https://github.com/albaaggbb/protein-expression-classification.git
cd protein-expression-classification

2. Place data3.csv and class3.csv in the data/ folder.

3. Run the notebook:

jupyter notebook notebooks/protein_expression_classification.ipynb

Results will be saved in the results/ folder.

├── data/
│   ├── data3.csv
│   └── class3.csv
├── notebooks/
│   └── protein_expression_classification.ipynb
├── results/
│   ├── class_distribution.png
│   ├── confusion_matrix_model_1.png
│   ├── confusion_matrix_model_2.png
│   ├── correlation_matrix.png
│   ├── data_normalized.csv
│   ├── learning_curve_model_1.png
│   ├── learning_curve_model_2.png
│   ├── summary_statistics.csv
│   └── protein_expression_classification.html
├── requirements.txt
├── .gitignore
└── README.md

Project developed by Alba Górriz.