Reg

Reg#

Index#

  1. Instantiate model class

  2. Define clock metadata

  3. Download clock dependencies

  4. Load features

  5. Load weights into base model

  6. Load reference values

  7. Load preprocess and postprocess objects

  8. Check all clock parameters

  9. Basic test

  10. Save torch model

  11. Clear directory

Let’s first import some packages:

[1]:

import os
import inspect
import shutil
import json
import torch
import pandas as pd
import pyaging as pya

Instantiate model class#

[2]:

def print_entire_class(cls):
    source = inspect.getsource(cls)
    print(source)

print_entire_class(pya.models.Reg)

class Reg(pyagingModel):
    def __init__(self):
        super().__init__()

    @staticmethod
    def _rank_average(values):
        """
        Assign average ranks (1-based) per vector, handling ties.
        """
        sorted_vals, sorted_idx = torch.sort(values)
        ranks = torch.empty_like(sorted_vals, dtype=values.dtype)

        n = values.numel()
        start = 0
        while start < n:
            end = start + 1
            while end < n and sorted_vals[end] == sorted_vals[start]:
                end += 1
            avg_rank = (start + end - 1) / 2.0 + 1.0
            ranks[sorted_idx[start:end]] = avg_rank
            start = end

        return ranks

    def preprocess(self, x):
        """
        Fill missing values with the global median then rank-normalize per sample.
        """
        median = torch.nanmedian(x)
        if torch.isnan(median):
            median = torch.tensor(0.0, device=x.device, dtype=x.dtype)
        x = torch.where(torch.isnan(x), median, x)

        ranked = torch.empty_like(x, dtype=x.dtype)
        for i in range(x.size(0)):
            ranked[i] = self._rank_average(x[i])

        return ranked

    def postprocess(self, x):
        """
        Add the REG intercept term after linear prediction.
        """
        intercept = self.postprocess_dependencies[0]
        return x + intercept


[3]:

model = pya.models.Reg()

Define clock metadata#

[4]:

model.metadata["clock_name"] = 'reg'
model.metadata["data_type"] = 'transcriptomics'
model.metadata["species"] = 'Homo sapiens'
model.metadata["year"] = 2025
model.metadata["approved_by_author"] = '✅'
model.metadata["citation"] = "Salignon, Jerome, et al. \"Pasta, an age-shift transcriptomic clock, maps the chemical and genetic determinants of aging and rejuvenation.\" bioRxiv (2025): 2025-06."
model.metadata["doi"] = "https://doi.org/10.1101/2025.06.04.657785"
model.metadata["research_only"] = None
model.metadata["notes"] = "Rank-normalized transcriptomic clock (Age in years)."

Download clock dependencies#

Download coefficient file#

[5]:

coeff_url = "https://raw.githubusercontent.com/bio-learn/biolearn/master/biolearn/data/REG.csv"
os.system(f"curl -L {coeff_url} -o REG.csv")

  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  315k  100  315k    0     0  1035k      0 --:--:-- --:--:-- --:--:-- 1037k

[5]:

0

Load features#

From CSV file#

[6]:

coeffs = pd.read_csv('REG.csv')
coeffs['feature'] = coeffs['GeneID']
coeffs['coefficient'] = coeffs['CoefficientTraining']

model.features = coeffs['feature'].tolist()

Load weights into base model#

From CSV file#

[7]:

weights = torch.tensor(coeffs['coefficient'].tolist()).unsqueeze(0)
intercept = torch.tensor([0.0])

Linear model#

[8]:

base_model = pya.models.LinearModel(input_dim=len(model.features))

base_model.linear.weight.data = weights.float()
base_model.linear.bias.data = intercept.float()

model.base_model = base_model

Load reference values#

[9]:

model.reference_values = [float("nan")] * len(model.features)

Load preprocess and postprocess objects#

[10]:

model.preprocess_name = "median_fill_and_rank_normalization"
model.preprocess_dependencies = None

[11]:

model.postprocess_name = "add_constant"
model.postprocess_dependencies = [140.272578432562]

Check all clock parameters#

[12]:

pya.utils.print_model_details(model)


%==================================== Model Details ====================================%
Model Attributes:

training: True
metadata: {'approved_by_author': '✅',
 'citation': 'Salignon, Jerome, et al. "Pasta, an age-shift transcriptomic '
             'clock, maps the chemical and genetic determinants of aging and '
             'rejuvenation." bioRxiv (2025): 2025-06.',
 'clock_name': 'reg',
 'data_type': 'transcriptomics',
 'doi': 'https://doi.org/10.1101/2025.06.04.657785',
 'notes': 'Rank-normalized transcriptomic clock (Age in years).',
 'research_only': None,
 'species': 'Homo sapiens',
 'version': None,
 'year': 2025}
reference_values: [nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan]... [Total elements: 8113]
preprocess_name: 'median_fill_and_rank_normalization'
preprocess_dependencies: None
postprocess_name: 'add_constant'
postprocess_dependencies: [140.272578432562]
features: ['ENSG00000196839', 'ENSG00000170558', 'ENSG00000133997', 'ENSG00000168060', 'ENSG00000101473', 'ENSG00000136754', 'ENSG00000113552', 'ENSG00000177485', 'ENSG00000136560', 'ENSG00000094631', 'ENSG00000108840', 'ENSG00000170248', 'ENSG00000153094', 'ENSG00000159921', 'ENSG00000165879', 'ENSG00000135451', 'ENSG00000142892', 'ENSG00000179776', 'ENSG00000167670', 'ENSG00000129484', 'ENSG00000041880', 'ENSG00000113361', 'ENSG00000141198', 'ENSG00000100284', 'ENSG00000013619', 'ENSG00000010017', 'ENSG00000105993', 'ENSG00000113810', 'ENSG00000182963', 'ENSG00000126261']... [Total elements: 8113]
base_model_features: None

%==================================== Model Details ====================================%
Model Structure:

base_model: LinearModel(
  (linear): Linear(in_features=8113, out_features=1, bias=True)
)

%==================================== Model Details ====================================%
Model Parameters and Weights:

base_model.linear.weight: [0.0003716579813044518, 6.28228226560168e-05, -8.446603897027671e-05, -0.00041367721860297024, -0.0002181005256716162, -0.0002126695035258308, 0.0005079449038021266, 9.584725921740755e-05, 0.0003452318487688899, -0.00011616084520937875, -2.2599540898227133e-05, 0.00019724950834643096, 0.00035635282984003425, -0.000530869874637574, -0.00013170151214580983, -5.169699579710141e-05, -0.000309494644170627, -5.409893856267445e-05, -0.0008479842217639089, -6.344888970488682e-05, 0.0010389157105237246, 2.4861426936695352e-05, 9.329131717095152e-05, -1.1981301213381812e-05, -9.546556248096749e-05, 8.20873974589631e-05, -0.00020033025066368282, -0.0009491293458268046, -0.00027923236484639347, -2.363449129916262e-05]... [Tensor of shape torch.Size([1, 8113])]
base_model.linear.bias: tensor([0.])

%==================================== Model Details ====================================%

Basic test#

[13]:

torch.manual_seed(42)
input = torch.randn(10, len(model.features), dtype=float)
model.eval()
model.to(float)
pred = model(input)
pred

[13]:

tensor([[106.5183],
        [-14.9794],
        [126.1043],
        [ 84.5967],
        [ 67.9882],
        [102.8186],
        [  6.6189],
        [ 94.9474],
        [ 88.0649],
        [-11.6583]], dtype=torch.float64, grad_fn=<AddBackward0>)

Save torch model#

[14]:

torch.save(model, f"../weights/{model.metadata['clock_name']}.pt")

Clear directory#

[15]:

# Function to remove a folder and all its contents
def remove_folder(path):
    try:
        shutil.rmtree(path)
        print(f"Deleted folder: {path}")
    except Exception as e:
        print(f"Error deleting folder {path}: {e}")

# Get a list of all files and folders in the current directory
all_items = os.listdir('.')

# Loop through the items
for item in all_items:
    # Check if it's a file and does not end with .ipynb
    if os.path.isfile(item) and not item.endswith('.ipynb'):
        os.remove(item)
        print(f"Deleted file: {item}")
    # Check if it's a folder
    elif os.path.isdir(item):
        remove_folder(item)

Deleted file: REG.csv
Contents