Zero-coverage control genes silently corrupt output with inf/NaN #11

Bug

When a BAM file has no reads covering one or more of the 20 housekeeping/control genes, the tool silently produces inf/NaN values in the output CSV with no warning.

Root cause

In sma.py, when H_ik[i,k] == 0 for any control gene, a chain of silent NumPy floating-point errors occurs:

H_ik[i,k] == 0.0
  → z_ik[i,k] = inf           (line 124: c_ix_sum / 0.0)
  → zmean_k[k] = inf          (line 127: sum of inf / n_bam)
  → theta_i[i] = nan           (line 128-129: inf/inf = nan)
  → pi_ij[i,j] = nan           (line 130-131: nan * anything = nan)

A single zero-coverage gene in a single sample corrupts all results for that sample. Furthermore, the inf values in z_ik leak into zmean_k, which is shared across all samples — so a single bad sample can degrade the statistics for every other sample in the batch.

Impact

• The user gets no error or warning — just silent NaN in the CSV output
• A single problematic BAM file can silently affect results for the entire batch
• This is especially likely with downsampled or targeted sequencing data where some control gene regions may have no coverage

Expected behavior

• Warn the user (to stderr) which samples have zero-coverage control genes and which genes are affected
• Set results to NaN explicitly for affected samples
• Exclude affected samples from shared statistics (zmean_k) so healthy samples are not corrupted