4fa15dade9f2174a602a28682d2f1013dba3ef78,src/pyscenic/binarization.py,,_derive_threshold,#,13

Before Change

    data = auc_mtx[regulon_name].values.reshape(-1, 1)
    gmm = mixture.GaussianMixture(n_components=2, covariance_type="full").fit(data)
    idx = np.argmax(gmm.means_)
    return max((gmm.means_[idx] - 1 * sqrt(gmm.covariances_[idx]))[0], 0)

def binarize(auc_mtx: pd.DataFrame) -> (pd.DataFrame, pd.Series):
    

After Change

    data = auc_mtx[regulon_name].values.reshape(-1, 1)
    gmm = mixture.GaussianMixture(n_components=2, covariance_type="full").fit(data)
    avgs = gmm.means_
    stds = np.sqrt(gmm.covariances_.reshape(-1, 1))

    // The threshold is based on the distribution with the highest mean and is defined as (mu - 2 x std)
    idx = np.argmax(avgs)
    threshold = max(avgs[idx] - 2 * stds[idx], 0)
    // This threshold cannot be lower than (mu + 2 x std) based on the distribution with the lowest mean.
    idx = np.argmin(avgs)
    lower_bound = avgs[idx] + 2 * stds[idx]

    return max(lower_bound, threshold)

def binarize(auc_mtx: pd.DataFrame) -> (pd.DataFrame, pd.Series):

Instances