ML-to-Predict-EE

Official repository of "A Machine Learning Framework for Predicting Entrapment Efficiency in Niosomal Particles".

Usage

First, you need to make your data into an n by m+1 matrix, where n (rows) is the number of data and m (columns) is the number of features, and the last column represents the desired output (entrapment efficiency or any feature you want).

Prepare Data

import numpy as np
from sklearn.utils import shuffle

data = ... # your data (n, m+1)

data = data.astype(np.float32) # convert all data dtype to float32
data = shuffle(data, random_state=41) # shuffle dataset

X, Y = data_[:,:-1], data_[:,-1] # seperate input and output (n, m+1) -> (n,m), (n,)
Y = np.expand_dims(Y, axis=-1) # add a dimension (n,) -> (n,1)

X = (X - X.min(axis=0)) / (X.max(axis=0) - X.min(axis=0) + 1e-8) # iput min-max normalization
Y = Y / 100 # output normalization

Use Trainer class

A class called ‍‍‍‍‍‍Trainer has been prepared for training models. Using Trainer, you can easily and without specialized knowledge specify the model, cost function, model hyperparameters, cost function hyperparameters, etc. Then train your model.

from einops import rearrange
from trainer import Trainer

trainer = Trainer(train_data=(X, Y), # your data
                  model_type="linear", # model_type (linear, polynomial, and dnn)
                  loss_type="gaussian", # loss_type(mae, mse, maem, and gaussian)
                  k_folds=5,
                  n_repeats=1, # number of repeat in each fold
                  epochs=1000, # number of training epoch in each fold
                  batch_size=512, # size of training batch
                  experiment_mode="normal", # training mode (normal or sensitivity)
                  epsilon=0.05, # epsilon in maem loss function (if you use other loss_type you don't need it)
                  in_features=X.shape[-1],
                  o_features=Y.shape[-1],
                  n_h_layers=6, # number of hidden layer for dnn (if you use other model_type you don't need it)
                  drop_rate=0.25, # dropout for dnn (if you use other model_type you don't need it)
                  h_units=256, # number of hidden unit for dnn (if you use other model_type you don't need it)
                  activation="relu", # type of activation function for dnn (if you use other model_type you don't need it)
                  degree=10, # degree of polynomial model (if you use other model_type you don't need it)
                  output_strict=True, # Limit the output to 1 or not
)

trainer() # Start training

Showing average metrics for K-fold cross-validation

results = rearrange(trainer.Train_Valid_Metric_Results, "f r m o -> (f r o) m")
results.mean(axis=0)[4:] # MAE, RMSE, R2, STD(if loss_type="gaussian") in order

Citation

If you find our code useful for your research, please consider citing:

@article{kashani2022machine,
  title={A Machine Learning Framework for Predicting Entrapment Efficiency in Niosomal Particles},
  author={Kashani-Asadi-Jafari, Fatemeh and Aftab, Arya and Ghaemmaghami, Shahrokh},
  journal={International Journal of Pharmaceutics},
  pages={122203},
  year={2022},
  publisher={Elsevier}
}