Cell Type Deconvolution¶

This dimension evaluates whether cell type composition inferred from bulk RNA-seq data is faithfully preserved in synthetic datasets. CIBERSORTx with the LM22 reference signature estimates the proportions of 22 human immune cell types.

Aitchison Distance¶

To quantify preservation of inferred immune cell compositions, we measured the similarity using the Aitchison distance, which is appropriate for compositional data

Compositional Center¶

Let \(x_{ij}\) denote the estimated fraction of cell type \(j\) in sample \(i\). The geometric mean of each cell type \(j\) across \(n\) samples:

\[g_j = \exp\!\left(\frac{1}{n}\sum_{i=1}^{n}\log x_{ij}\right)\]

CLR Transformation¶

The Centered Log-Ratio transformation of the compositional center vector:

\[\text{CLR}(g_j) = \log\!\left(\frac{g_j}{\left(\prod_{l=1}^{k} g_l\right)^{1/k}}\right)\]

Distance and Score¶

\[d_A = \sqrt{\sum_{j=1}^{k}\bigl(\text{CLR}(g_j^{\text{original}}) - \text{CLR}(g_j^{\text{synthetic}})\bigr)^2}\]

Converted to a score for easier comparison (higher = better):

\[\text{Score} = \exp(-d_A)\]

where the analysis is constrained to the top \(k = 10\) most abundant cell types.

Code Example¶

The cell_deconvolution module in synomicsbench provides the Aitchison distance calculation:

from synomicsbench.metrics.narrow_utility.cell_deconvolution import (
    aitchison_distance,
    aitchison_score,
)

cell_types = ['T cells CD8', 'T cells CD4 mem. resting', 'Plasma cells']
# Compute Aitchison distance and score
d = aitchison_distance(ori_deconvolution, syn_deconvolution, cell_types)
score = aitchison_score(d)

print(f"Aitchison Distance: {d:.4f}")
print(f"Aitchison Score: {score:.4f}")

The ori_deconvolution and syn_deconvolution objects should be DataFrames where rows are samples and columns represent cell type fractions:

Patient	B cells naive	B cells memory	Plasma cells	T cells CD8	T cells CD4 mem. resting
Patient1	0.0637	0.1255	0.0985	0.0500	0.1437
Patient10	0.1348	0.0000	0.1443	0.1457	0.4914
Patient100	0.0949	0.1930	0.6228	0.0866	0.0000
Patient102	0.0502	0.1129	0.2339	0.1211	0.4372
Patient105	0.0080	0.0488	0.0218	0.1817	0.2385