BASEN: Preprocessing and QC Record

This page documents preprocessing and quality-control procedures that are summarized only briefly in the main report. It includes basecalling, read-quality assessment, read-length filtering, host-read removal, and final quality filtering for the three mock communities: Microbial (D6300), Gut (D6331), and Oral (D6332).

1. Preprocessing

1. Basecalling
• MinKNOW did not include the most recent Dorado basecaller used for this analysis. POD5 files that could not be basecalled during sequencing were retained in pod5_skip/.
• All POD5 files from pod5/ and pod5_skip/ across the 18 sequencing runs were therefore re-basecalled on high-performance computing clusters using Dorado v2.1.2, with the same software version and basecalling model applied to every run.
• Basecalling model: dna_r10.4.1_e8.2_400bps_sup@v5.2.0


2. Quality assessment

Sequencing quality was assessed at both the run and read levels using:

• Dorado Run Analyzer: sequencing-run summaries derived from MinKNOW reports, including read length, read quality, and pore activity.
• Detailed MultiQC results: summaries generated by sequencing plate, barcode, and read-length interval to evaluate read-quality patterns and support the choice of length-filtering criteria.


3. Read quality control

Inspection of read quality across length intervals supported the use of a single read-length rule for all datasets:

1. Short reads were retained. Reads shorter than the Kraken2 k-mer length (35 bp) contain no complete k-mers and therefore remain unclassified by Kraken2.
2. Reads longer than 5,800 bp were trimmed to 5,800 bp. This threshold was selected because per-base quality declined beyond this length in Microbial-05 (D6300, run 5, barcode05), where mean per-base Q decreased from 34.7 at 5,800 bp to 24.1 at 6,000 bp. The same threshold was applied to all datasets as a conservative and consistent preprocessing rule.
3. Ultra-long reads (≥ 100,000 bp) were removed. These reads showed substantially lower mean quality, with mean Q values of 2.8–3.2 across the three mock communities.

Following length processing, reads mapping to the human reference genome GRCh38 were removed using Minimap2 v2.30. Reads with a Dorado mean Q score below 20 were then excluded.

4. Reads retained at each preprocessing step

Read counts were recorded for each run after basecalling, ultra-long-read removal, length trimming, host-read removal, and final quality filtering (table).

Run	Barcode	Basecalled	≥ 100,000 bp (removed)	Trimmed to 5,800 bp	Human	Host-free	Q ≥ 20
Microbial-01	barcode21	1,385	1	27	3	1,381	327
Microbial-02	barcode23	656	0	31	3	653	247
Microbial-03	barcode01	5,195	0	80	15	5,180	1,223
Microbial-04	barcode03	6,222	0	251	17	6,205	1,725
Microbial-05	barcode05	541	0	5	2	539	105
Microbial-06	barcode09	58,109	1	2,393	197	57,911	11,351
Gut-01	barcode15	8,245	3	226	33	8,209	1,057
Gut-02	barcode17	2,624	0	39	4	2,620	375
Gut-03	barcode19	1,683	0	29	3	1,680	365
Gut-04	barcode07	5,353	0	282	4	5,349	1,022
Gut-05	barcode11	30,108	2	1,215	38	30,068	7,688
Gut-06	barcode13	18,862	0	1,162	25	18,837	2,860
Oral-01	barcode05	263,381	11	18,040	188	263,182	165,929
Oral-02	barcode07	198,084	16	10,501	173	197,895	111,559
Oral-03	barcode09	172,739	5	5,555	99	172,635	85,055
Oral-04	barcode11	147,603	4	4,701	91	147,508	75,026
Oral-05	barcode13	53,383	1	2,304	21	53,361	30,780
Oral-06	barcode15	69,130	0	2,368	33	69,097	36,485

The Q ≥ 20 reads constitute the final input to Kraken2. The median read length of these reads was also used as the corresponding Bracken read-length parameter for each run.