app.py

#@title Lauch the Interface
#@markdown Run to launch the interface.
#@markdown 1. Enter the protein sequence.
#@markdown 2. Specify the start and end index (inclusive) of the domain for which you would like to predict mutations (note that indexing starts at 1).
#@markdown 3. Select the number of tokens you would like the model to predict for each position.
#@markdown 4. Click 'Submit'.
#@markdown 5. Click 'Download Outputs' to download the zip file.

def process_sequence(sequence, domain_bounds, n):
    start_index = int(domain_bounds['start'][0]) - 1
    end_index = int(domain_bounds['end'][0])

    top_n_mutations = {}
    all_logits = []

    for i in range(len(sequence)):
          if start_index <= i <= (end_index - 1):
              masked_seq = sequence[:i] + '<mask>' + sequence[i+1:]
              inputs = tokenizer(masked_seq, return_tensors="pt", padding=True, truncation=True, max_length=2000)
              inputs = {k: v.to(device) for k, v in inputs.items()}
              with torch.no_grad():
                  logits = model(**inputs).logits
              mask_token_index = torch.where(inputs["input_ids"] == tokenizer.mask_token_id)[1]
              mask_token_logits = logits[0, mask_token_index, :]

              # Define amino acid tokens
              AAs_tokens = ['L', 'A', 'G', 'V', 'S', 'E', 'R', 'T', 'I', 'D', 'P', 'K', 'Q', 'N', 'F', 'Y', 'M', 'H', 'W', 'C']
              all_tokens_logits = mask_token_logits.squeeze(0)
              top_tokens_indices = torch.argsort(all_tokens_logits, dim=0, descending=True)
              top_tokens_logits = all_tokens_logits[top_tokens_indices]
              mutation = []
              # make sure we don't include non-AA tokens
              for token_index in top_tokens_indices:
                  decoded_token = tokenizer.decode([token_index.item()])
                  if decoded_token in AAs_tokens:
                      mutation.append(decoded_token)
                      if len(mutation) == n:
                          break
              top_n_mutations[(sequence[i], i)] = mutation

              # collecting logits for the heatmap
              logits_array = mask_token_logits.cpu().numpy()
              # filter out non-amino acid tokens
              filtered_indices = list(range(4, 23 + 1))
              filtered_logits = logits_array[:, filtered_indices]
              all_logits.append(filtered_logits)


    token_indices = torch.arange(logits.size(-1))
    tokens = [tokenizer.decode([idx]) for idx in token_indices]
    filtered_tokens = [tokens[i] for i in filtered_indices]

    all_logits_array = np.vstack(all_logits)
    normalized_logits_array = F.softmax(torch.tensor(all_logits_array), dim=-1).numpy()
    transposed_logits_array = normalized_logits_array.T

   # Plotting the heatmap
    x_tick_positions = np.arange(start_index, end_index, 10)
    x_tick_labels = [str(pos + 1) for pos in x_tick_positions]

    plt.figure(figsize=(15, 8))
    plt.rcParams.update({'font.size': 18})

    sns.heatmap(transposed_logits_array, cmap='plasma', xticklabels=x_tick_labels, yticklabels=filtered_tokens)
    plt.title('Token Probability Heatmap')
    plt.ylabel('Token')
    plt.xlabel('Residue Index')
    plt.yticks(rotation=0)
    plt.xticks(x_tick_positions - start_index + 0.5, x_tick_labels, rotation=0)

    # Save the figure to a BytesIO object
    buf = BytesIO()
    plt.savefig(buf, format='png', dpi = 300)
    buf.seek(0)
    plt.close()

    # Convert BytesIO object to an image
    img = Image.open(buf)

    original_residues = []
    mutations = []
    positions = []

    for key, value in top_n_mutations.items():
        original_residue, position = key
        original_residues.append(original_residue)
        mutations.append(value)
        positions.append(position + 1)

    df = pd.DataFrame({
        'Original Residue': original_residues,
        'Predicted Residues': mutations,
        'Position': positions
    })
    df.to_csv("predicted_tokens.csv", index=False)
    img.save("heatmap.png", dpi = 300)
    zip_path = "outputs.zip"
    with zipfile.ZipFile(zip_path, 'w') as zipf:
        zipf.write("predicted_tokens.csv")
        zipf.write("heatmap.png")


    return df, img, zip_path

demo = gr.Interface(
    fn=process_sequence,
    inputs=[
        gr.Textbox(label="Sequence", placeholder="Enter the protein sequence here"),
        gr.Dataframe(
            headers=["start", "end"],
            datatype=["number", "number"],
            row_count=(1, "fixed"),
            col_count=(2, "fixed"),
            label="Domain Bounds"
        ),
        gr.Dropdown([i for i in range(1, 21)], label="Top N Tokens"),
    ],
     outputs=[
        gr.Dataframe(label="Predicted Tokens (in order of decreasing likelihood)"),
        gr.Image(type="pil", label="Heatmap"),
        gr.File(label="Download Outputs"),
    ],
)
if __name__ == "__main__":
    with suppress_output():
        demo.launch(show_error=False, debug=False)