app.py

import os
os.system("pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu")
os.system("pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.12.0+cpu.html")
import gradio as gr
from glycowork.ml.processing import dataset_to_dataloader
import numpy as np
import torch
import torch.nn as nn
from glycowork.motif.graph import glycan_to_nxGraph
import networkx as nx
import pydot
# import pygraphviz as pgv


class EnsembleModel(nn.Module):
    def __init__(self, models):
        super().__init__()
        self.models = models
        
    def forward(self, data):
      # Check if GPU available
      device = "cpu"
      if torch.cuda.is_available():
        device = "cuda:0"
      # Prepare data
      x = data.labels.to(device)
      edge_index = data.edge_index.to(device)
      batch = data.batch.to(device)
      y_pred = [model(x,edge_index, batch).cpu().detach().numpy() for model in self.models]
      y_pred = np.mean(y_pred,axis=0)[0]
      return y_pred
  
class_list=['Amoebozoa', 'Animalia', 'Bacteria', 'Bamfordvirae', 'Chromista', 'Euryarchaeota', 'Excavata', 'Fungi', 'Heunggongvirae', 
            'Orthornavirae', 'Pararnavirae', 'Plantae', 'Proteoarchaeota', 'Protista', 'Riboviria']

model1 = torch.load("model1.pt", map_location=torch.device('cpu'))
model2 = torch.load("model2.pt", map_location=torch.device('cpu'))
model3 = torch.load("model3.pt", map_location=torch.device('cpu'))
model4 = torch.load("model4.pt", map_location=torch.device('cpu'))

def fn(glycan, model):
    # Draw graph
    graph = glycan_to_nxGraph(glycan)
    node_labels = nx.get_node_attributes(graph, 'string_labels')
    labels = {i:node_labels[i] for i in range(len(graph.nodes))}
    graph = nx.relabel_nodes(graph, labels)
    graph = nx.drawing.nx_pydot.to_pydot(graph)
    graph.set_prog("dot")
    graph.write_png("graph.png")
    # write_dot(graph, "graph.dot")
    # graph=pgv.AGraph("graph.dot")  
    # graph.layout(prog='dot')
    # graph.draw("graph.png")
    # Perform inference
    if model == "No data augmentation":
      model_pred = model1
      model_pred.eval()
    elif model == "Ensemble":
      model_pred = model3
      model_pred.eval()
    elif model == "Bootstrap Ensemble":
      model_pred = model4
      model_pred.eval()
    else:
      model_pred = model2
      model_pred.eval()
    
    glycan = [glycan]
    label = [0]
    data = next(iter(dataset_to_dataloader(glycan, label, batch_size=1)))
    
    if model == "Ensemble":
        pred = model_pred(data)
    else:
        device = "cpu"
        x = data.labels
        edge_index = data.edge_index
        batch = data.batch
        x = x.to(device)
        edge_index = edge_index.to(device)
        batch = batch.to(device)
        pred = model_pred(x,edge_index, batch).cpu().detach().numpy()[0]
    
    pred = np.exp(pred)/sum(np.exp(pred)) # Softmax 
    pred = [float(x) for x in pred]
    pred = {class_list[i]:pred[i] for i in range(15)}
    return pred, "graph.png"


demo = gr.Interface(
    fn=fn,
    inputs=[gr.Textbox(label="Glycan sequence"), gr.Radio(label="Model",choices=["No data augmentation", "Random node deletion", "Ensemble", "Bootstrap Ensemble"])],
    outputs=[gr.Label(num_top_classes=15, label="Prediction"), gr.Image(label="Glycan graph")],
    allow_flagging="never",
    title="SweetNet demo",
    examples=[["GlcOSN(a1-4)GlcA(b1-4)GlcOSN(a1-4)GlcAOS(b1-4)GlcOSN(a1-4)GlcOSN", "No data augmentation"],
    ["Man(a1-2)Man(a1-3)[Man(a1-3)Man(a1-6)]Man(b1-4)GlcNAc(b1-4)GlcNAc", "Random node deletion"],
    ["GlcNAc(b1-7)LDManHep(a1-6)Glc(a1-2)Glc(a1-3)[Gal(a1-6)]Glc(a1-3)[LDManHep(a1-7)]LDManHepOP(a1-3)LDManHepOP(a1-5)[Kdo(a2-4)]Kdo", "Ensemble"]]
)
demo.launch(debug=True)