dlim 1

Model genotype to fitness map

Overview

D-LIM (Direct-Latent Interpretable Model) is a neural network that enhances genotype-fitness mapping by combining interpretability with predictive accuracy. It assumes independent phenotypic influences of genes on fitness, leading to advanced accuracy and insights into phenotype analysis and epistasis. The model includes an extrapolation method for better understanding of genetic interactions and integrates multiple data sources to improve performance in low-data biological research.

Requirements

The implementation has been tested on a Linux system for: Python 3.10.9; Pytorch 2.0.1; numpy 1.23.5; pandas 2.0.2.

Installation

Install the package from Pypi:

pip install dlim

Or install it from the sources:

pip install .

Usage

The code snippet bellow shows how to use fit D-LIM for fitness prediction.

Load packages

from dlim.model import DLIM 
from dlim.dataset import Data_model
from dlim.api import DLIM_API
from scipy.stats import pearsonr, spearmanr
from sklearn.metrics import r2_score
from numpy import mean
from numpy.random import choice
import pandas as pd
from tqdm import tqdm  
import matplotlib.pyplot as plt 
import numpy as np 

Read the data, there are 2 genes (or variables) here

df_data = pd.read_csv("../data/data_env_1.csv", sep = ',', header = None)
data = Data_model(data=df_data, n_variables=2)

Create DLIM model

# Here, we used 2 latent phenotype, the data has 37 possible mutations.
# For D-LIM, we use here 1 hidden layer of 32 neurons.
model = DLIM(n_variables = 2, hid_dim = 32, nb_layer = 0)
dlim_regressor = DLIM_API(model=model, flag_spectral=True)

Split data into training and validation

train_id = choice(range(data.data.shape[0]), int(data.data.shape[0]*0.7), replace=False)
val_id = [i for i in range(data.data.shape[0]) if i not in train_id]
train_data = data.subset(train_id)
val_data = data.subset(val_id)

Train model

# We train the model with a learning rate of 1e-2 for 300 steps with batch size
# 16 and regularization of 1e-2
losses = dlim_regressor.fit(train_data, lr = 1e-2, nb_epoch=300, batch_size=32, emb_regularization=1e-2)

Prediction on validation data

# Now, we compute the validation prediction
fit, var, _  = dlim_regressor.predict(val_data.data[:,:-1], detach=True) 

score = pearsonr(fit.flatten(), val_data.data[:, [-1]].flatten())[0]
print(score)

Plot and get landscape

# Here, we plot the trained landscape
fig, (bx, cx, dx) = plt.subplots(1, 3, figsize=(6, 2))
dlim_regressor.plot(bx, data)

for xx in [bx, cx, dx]:
    for el in ["top", "right"]:
        xx.spines[el].set_visible(False)

# Plot the a00verage curve
print(pearsonr(lat_a[:, 0], data[:, -1]))
cx.scatter(lat_a[:, 0], data[:, -1], s=5, c="grey")
dx.scatter(lat_a[:, 1], data[:, -1], s=5, c="grey")
cx.set_ylabel("F")
dx.set_xlabel("$\\varphi^1$")
cx.set_xlabel("$\\varphi^2$")
plt.tight_layout()
plt.show()

<p align="center"><img src="https://github.com/LBiophyEvo/D-LIM-model/blob/main/reproducibility/img/spec_harry_env_1.png" /></p>

More details of the usage and the installation can be found in doc:

• Read the docs of D-LIM

Data download:

• Datasets

None

Reproduction of the manuscript

Figures and analyses of the manuscript can be found in File reproducibility.

License

MIT