Direct RNA sequencing dwell to millisecond conversion

[XqKang]

Hi Remora Development Team,

First, I'd like to thank you for developing Remora. It has been an invaluable tool for my direct RNA sequencing data analysis.

I am currently using Remora to compare modified versus canonical RNA samples and am preparing a manuscript for publication. For the sake of accuracy, I would appreciate clarification on the precise definitions of a few key metrics.

1. dwell Metric: Unit and Conversion to Time
   My understanding is that the dwell value represents the number of raw signal measurements (samples) for each base. To accurately label figures in our publication, could you please confirm the correct method for converting this sample count into milliseconds (ms)? We assume the conversion requires the experiment's sampling frequency (e.g., time_ms = samples / sampling_rate_hz * 1000), but any official guidance would be greatly appreciated.

2. trimmean Metric: Unit and Normalization Method
   We are using the trimmean metric with the default signal_type="norm". Am I correct in assuming that since the signal is normalized, the resulting trimmean value is a unitless, dimensionless quantity? Could you please confirm if the default "norm" signal corresponds to a standard Z-score normalization (Z=(x−mu)/sigma)?

Here is the script i use to extract the dwell and trimmean:
samples_metrics, all_bam_reads = io.get_ref_reg_samples_metrics(
ref_reg,
[(pod5_dr, can_bam_fh), (pod5_dr, mod_bam_fh)],
metric="dwell_trimmean_trimsd",
sig_map_refiner=sig_map_refiner,
reverse_signal=True,
max_reads=100000
)

[marcus1487]

Thanks for your question, and we’re glad to hear Remora has been helpful for your RNA modification work.

Dwell metric:
You’re correct. The dwell value is the number of raw signal samples (ADC measurements) assigned to each base. To convert this to milliseconds, divide by the sampling rate (in Hz) and multiply by 1000: time_ms = dwell_samples / sampling_rate_hz * 1000

For example, with a 4000 Hz sampling rate, a dwell of 120 samples equals 30 ms.

Trimmean metric:
With signal_type="norm", the trimmean value is unitless. The default normalized signal is based on Z-score normalization: z = (x - mean) / std. The exact normalization depends on the sig_map_refiner settings. The example notebooks in the Remora repo describe the normalization process in more detail. If you’d like clarification for a specific refiner configuration, feel free to share it here.