Short tandem repeats (STRs) in human genome are specific DNA loci that can have high degree of polymorphisms in the number and forms of repeat motifs among individuals. This feature makes STRs a kind of genetic markers well-suited for individual identification [1,2], cell line authentication [3,4], as well as relationship inferring [5–7]. In Short Tandem Repeat DNA Internet Database maintained by the National Institute of Standards and Technology (NIST) [8], more than 70 STR genetic markers, typically with 19 (D10S1248) -114 (FGA) alleles in human populations, are chosen as genetic markers with desirable features as high heterozygosity, regular repeat unit, distinguishable alleles and robust amplification. These STRs have been utilized for decades by forensic laboratories across the globe and have formed the basis of million- or even 10-million-scale human DNA databases for public security purposes in many countries. In the realm of biomedical research, profiling STR for cell line authentication plays a critical role in ensuring scientific reproducibility and preventing misidentification as well as cross-contamination of cell lines[9,10]. To establish standardized practices for cell line authentication using STRs, the International Cell Line Authentication Committee (ICLAC, https://iclac.org) integrates numerous 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 online databases and search tools, including the Cellosaurus knowledge resource [11], DSMZ STR profile database[12], CLIMA database [13], CLASTER [14].

Conventional method for STR genotyping is the use of capillary electrophoresis (CE)-based length analysis. Besides, massively parallel sequencing (MPS) has also been validated for STR profiling, and it provide a larger battery of STRs in a run than the CE-based method. However, the majority of instruments of Sanger sequencing and MPS are heavy (50-186 kg), costly (50k-180k US dollars), and have high environmental requirements as constant temperature and vibration free. Thus, Sanger sequencing and MPS systems are usually maintained by well-equipped laboratories or facilities, and samples are transported for analysis, which might needs days to obtain STR typing reports.

A more rapid and affordable solution of STRs profiling thus has significance for relevant research and applications. Nanopore sequencing provides a cheap and real-time sequencing approach. The devices for nanopore sequencing, such as MinION Mk1B/Mk1c (Oxford Nanopore Technologies, Oxford, UK) and the recently released QNome-3841 (Qitan Technology, Chengdu, China), are exceptionally portable and considerably cheaper than NGS and Sanger sequencing systems. Moreover, these portable nanopore sequencing devices are capable of generating much longer reads than Sanger sequencing and NGS systems, and had a throughput in the Gbp range during a single run. Therefore, nanopore sequencing has been applied in various on-site sequencing applications as in-field genomic surveillance of pathogens [15,16] and biodiversity investigations [17,18]. Due to the portability and high throughput, Zaaijer et al.[19] reported an interesting pilot study that utilized MinION for whole genome sequencing to rapidly profile SNPs and authenticate human cell lines for biological researches. 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84

However, despite these advantages, nanopore sequencing data are noisier than these generated by Sanger sequencing and MPS. Unlike the cases of STR expansion disorders causing diseases, in which pathogenic alleles remarkably differ from benign alleles in sizes [20–22], alleles of STR genetic markers for individual or cell-line identity identification are usually of a 100 – 300 nt length and their repeat units are of 3- to 5-nt. Genotypes of STR genetic markers are defined by the exact counts of their repeats. Many STR genetic markers consist of multiple (compound STRs and complex STRs) and units are similar within in one STR and its flanking regions. For instance, the repeat structure of a Combined DNA Index System (CODIS) core STR locus D3S1358 is TCTA[TCTG]n[TCTA]m, in which n and m denotes repeat numbers. Moreover, some repeat units of STRs contain or form 3- to 4-nt homopolymers or form homopolymers in repeats, as STR Penta D with a repeat structure of [AAAGA]m, which are particularly error-prone, systematically, in nanopore sequencing [23].

Thus, the improvement of highly accurate typing method for genetic STR markers for nanopore sequencing is challenging [24–27]. To tackle this issue , developing algorithms and workflows specifically tailored for typing of genetic STR markers, considering repeat structures of specific STR loci, might be feasible strategy. In our previous study [23], we exhibited that conventional genotyping tools for nanopore sequencing as repeatHMM [21] are not suitable, and a simple customized workflow that integrated the classic Smith-Waterman algorithm for repeat unit re-alignment provided a significant improvement in the accuracy of STR typing. More than 32 STRs could be consistently and correctly typed in a validation set involving 31 individuals, and it is a prove-of-concept study of employing nanopore sequencing in human identity testing and cell-line authentication. Recently, Hall et al. [27] and Tytgat et al. [24] and the very recently 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105

Jidong et al [28], used a different and straight strategy that includes all available sequences of alleles for alignment of nanopore sequencing reads, and to select the alleles with sequencing reads for STR genotyping. In their validation sets, they report a higher STR typing accuracies than our previous method, and it further indicates the promising application of nanopore sequencing in this field. However, this strategy highly depends on the available sequences, many of which are unknown for these highly diverse STRs and might lead to error typing. On the other side, larger validation sample sets are also needed to example validate the STR typing methods. We aim to develop a method for accurate STR genotyping in applications as individual identification and cell-line authentication. We expect that the method can fully consider the characteristics of these STR markers and have high flexibility, i.e., independent of allele sequences databases that lack allele information. In this study, we propose NASTRA, a tool with clustering-based structure-aware algorithm , rather than allele-dependent alignment, for inference of repeat structure for STR genotyping. We includes a validation set involving 76 individuals and 8 cell-lines, and tested with different chemistries (R9 and R10 flow cells) as well as STR amplification methods prior to sequencing. Meanwhile, the validation based on allele-dependent methods were conducted and NASTRA turned out to have better performance in both accuracy and sequencing data requirement.

In this study, we firstly illustrated the overview of NASTRA. And then demonstrated how to find the optimal thresholds: supporting reads number (SN) and the ratio of supporting reads number (SNR), which is pivotal for STR genotype calling. Following this, we conducted an extensive 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 evaluation of NASTRA, comparing it with a typical alignment-based method named STRspy and investigating the characteristics of genotyping errors. We then conducted a real-world scenario testing using eight human cell lines. Lastly, we summarized the runtime of NASTRA. Throughout these processes, we leveraged the MiSeq FGx™ Forensic Genomics System to carry out STR genotype calling for ground truth construction. Specifically, we amplified 54 STR loci and 46 STR loci across 88 DNA samples using Verogen’s Foren Seq Mix A kit and Promega’s PowerSeq 46GY kit and sequenced amplicons with R9.4.1 and R10.3 flowc ells (Figure 1a).

In this study, we developed a novel computational approach for STR genotyping, named NASTRA. NASTRA comprises two key algorithms: read clustering and repeat structure inference. Recognizing that sequencing errors can occur randomly occurs in each amplicon read, NASTRA utilizes read clustering to mitigate the impact of these errors and found the candidate allele sequences of the STR. Once the candidate allele sequences are obtained, structure-aware algorithm will reconstruct the repeat structures of alleles based on the known repeat motifs from fact sheets. This process transforms the allele sequences into bracket sequences, where incomplete units or SNPs are readily discernible. Notably, in contrast to alignment-based methods, NASTRA operates independently of allele reference database and thus won’t fail to 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 genotype STR resulting from the lack of allele information in the database. Additionally, we have applied NASTRA on MiSeq FGx data for ground truth construction, demonstrating it can also be used for NGS data.

To comprehensively evaluate the performance of NASTRA, we carried out several assessment procedures. We first created a ground truth dataset of 88 DNA samples using the MiSeq FGx system, which is a widely accepted technology for forensic applications. Then, amplification and sequencing experiments were carried out across these samples using various amplification kits and flow cells. To investigate the optimal thresholds for NASTRA and its robustness under different sequencing durations, we expanded the training and test sets using NanoTime. Our performance evaluation encompassed ForenSeq data, PowerSeq data, and a real-world test involving eight standard DNA cell lines. For performance comparison, we exclusively utilized the alignment-based STRspy and did not employ well-known repeat quantification tools such as repeatHMM [21], STRique [22], and DeepRepeat [20], as they are primarily tailored for detecting repeat expansion disorders rather than forensic-level STR genotyping. As demonstrated in our previous study [23], repeatHMM didn’t perform well. However, DeepRepeat presents an intriguing possibility for utilizing current signals in STR genotyping. Exploring transfer learning with our data could be a promising avenue. In summary, the results demonstrated that NASTRA exhibited exceptional performance in terms of accuracy and speed, making it a promising method for real-world applications.

However, NASTRA does have several limitations. Firstly, it's worth noting that our performance comparison was conducted using only one alignment-based method, and the reference database for this method is still undergoing updates. Therefore, this evaluation may not provide a fully 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 objective assessment but rather serves as an illustrative example for reference in terms of accuracy and runtime. Secondly, NASTRA is unable to genotype incomplete repeats. While we have demonstrated that incomplete repeat units can be detected in the output allele, NASTRA does not incorporate the digital information into its allele calls. Thirdly, NASTRA’s read clustering process carries the potential risk of erroneously classifying two alleles as originating from a single allele. This may occur due to the tolerance of a single nucleotide polymorphism (SNP) or a one-base gap when dealing with two similar alleles (differing by just one-base). Fourthly, we have observed that the length of amplicons can impact performance based on our testing with ForenSeq and PowerSeq data. Specifically, the flanking regions of these STRs, including their prefix and suffix, can be quite similar, potentially leading to incorrect trimming. NASTRA requires amplicons with long flanking regions (more than 30 bp) to ensure the accuracy of STR genotyping. Finally, NASTRA relies on base sequences as input, and its performance is therefore dependent on the base calling model. Given that STRs are a type of complex genomic region, the development of an effective base calling model for STRs is necessary.

NASTRA is an innovative computational approach for STR genotyping using nanopore sequencing data, employing a structure-aware algorithm instead of traditional alignment to a reference database. Through tests and comparisons conducted on various flow cells and amplification kits, NASTRA demonstrated good performance in terms of genotyping accuracy and speed. Although not all STR loci obtained correct genotype calls, NASTRA exhibited robustness in certain specific loci, demonstrating its significant potential for human identification and cell line authentication 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 using nanopore sequencing platforms. Further improvements of NASTRA through validation studies with larger sample sizes are necessary.

Sample collection and DNA extraction A total of 76 blood samples were collected from 76 unrelated anonymized Han Chinese volunteers, including 38 females (ID: F01-F38) and 38 males (ID: M01-M38). Among them, 30 samples (F01-F15 and M01-M15) were used in our pervious study [23]. The rest 46 samples (F16-F38 and M16-F38) were newly collected. Genomic DNA of newly collected 46 samples were extracted using PureLink genomic DNA kit (Invitrogen, USA). In order to perform a real-time scenario tesing, we collected 8 control DNA samples including 3 forensic DNA controls (2800M from Promega, 9947A and 9948 from Origene), and 5 cell line standards (NA12878, NA24143, NA24149, NA24694, and NA24695 from Coriell Institute). All DNA samples were quantified on the Qubit 3.0 Fluorometer (Invitrogen) before amplification.

STR profiling using MiSeq FGx system Newly collected DNA samples (F16-F30, M16-M30) and 8 control DNA samples were amplified using the ForenSeq DNA Signature Prep Kit (DNA Primer Mix A). In the meantime, 2800M were amplified as positive control using DNA Primer Mix B, which covers all STR loci in Mix A kit. To obtain sufficient amplicons for Illumina MiSeq FGx sequencing and Nanopore sequencing, 50 ng of template rather than recommended 1 ng were used as input. PCR amplification were performed on ProFlex™ 3x32-Well PCR System (Thermo Fisher). 347

analytical threshold (1.5%), interpretation thresho ld (4.5%) and stutter filter (specific to each 348

locus). 341 342 343 344 345 346 349 350 351 352 353 354 355 356 357 358 359 360

All purified amplified libraries were divided into 2 shares, which were used for both MiSeq FGx sequencing and Nanopore sequencing. Pooled libraries were obtained combining equal volumes (2 μL) of normalized library (10 nmol) and diluted into 4 nmol. Finally, 7 pmol of pooled libraries were sequenced on the MiSeq FGx desktop sequencer. The above experimental steps were carried out according to the manufacturer’s protocol. Genotype calling of STR loci was performed using ForenSeq Universal Analysis Software, UAS (v1.3.6897; Verogen), with Verogen’s default Nanopore sequencing for ForenSeq amplicons The concentration of PCR products was quantified with Qubit 3.0 fluorometer and Qubit dsDNA HS assay kit according to manufacturer’s instruction, before library preparation. Nanopore libraries were prepared with Ligation Sequencing Kit (SQK-LSK109). Briefly, 0.2 pmol purified DNA per sample was processed using NEBNext Ultra II End repair/dA-tailing Module(E7546). The samples were multiplexed using Native Barcoding Expansion 1-12 (PCR-free) and Native Barcoding Expansion 13-24 (PCR-free). Adapter was ligated to the pooled libraries with NEBNext Quick Ligation Module (E6056). Finally, 0.05 pmol libraries were loaded into the R9.4.1 flowcell and sequencing was performed with MinKNOW. In general, 48 DNA samples and 8 control DNA samples were sequenced in 5 runs with a sequencing duration beyond 24 hours. Nanopore sequencing for PowerSeq amplicons According to the manufacturer’s instruction, 48 DNA sample were selected and amplified using the PowerSeq™ 46GY System (Promega, WI, USA). 1 ng template DNA of each sample was used 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 for PCR amplification with 29 cycles. The PCR products were clean-up with AmPure XP Bead (Beckman Coulter, Indianapolis, USA) and DNA concentration. The library preparation process is the same as that was described above. Finally, 0.05 pmol libraries were loaded into the R10.3 flowcell and sequencing was performed using MinKNOW. In general, 48 DNA samples were sequenced in 2 runs with a sequencing duration beyond 24 hours.

Bioinformatic analysis for nanopore sequencing data We performed the basecalling process by using Guppy v6.3.4 with high-accuracy model. Then, reads were aligned to the human reference genome (assembly GRCh37, hg19) using Minimap2 v2.17-r941 [30,31]. Finally, SAMtools v1.6 [ 32 ] were employed to convert SAM files into BAM format, which were used as input of NASTRA. Downsampled data were generated using NanoTime, which is available at https://github.com/renzilin/NanoTime.

Pairwise alignment with affine-gap penalty We used parasail-python [29] (https://github.com/jeffdaily/parasail-python) to perform pairwise alignment with affine-gap penalty in read trimming and clustering steps. Because gap open and extension penalties are required to be positive integers, we specified the match score of 75, the mismatch penalty of 90, gap open penalty of 75, and gap extension penalty of 10. Consequently, the alignment prefers to 2 continuous gaps rather than 1 mismatch. If the number of continuous gaps exceeds 2, the alignment prefers 1 mismatch.

Recursive algorithm for repeat structure inference To obtain the number of repeat units and finish the allele call without alignment, we developed a recursive algorithm to infer the repeat structure by searching repeat motif in allele sequence 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 according to the STRBase fact sheets. The algorithm works as following: 1. for each motif, exact match is used to search the segments consisting of a single type of motif; 2. we masked one segment at a time and repeat step 1; 3. remove duplicated masked reads and redo step 2 until no new segment is found; 4. if masked reads converge to the only one, then this is the optimal inference of repeat structure. The schematic representation of algorithm is shown in Figure 6.

Ethics approval and consent to participate Our study was approved by the ethics committee of Shanxi Medical University (no. 2020GLL031). Written informed consent was obtained from all participants and all authors.

Not applicable. 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426

ZR developed the computational method, conducted all data analysis and drafted the manuscript. JZ and TY conducted all wet laboratory experiments, including DNA extraction, amplification, and MiSeq FGx and Nanopore sequencing. YZ conducted the test of computational test. PS revised the manuscript. JX revised the manuscripts. JY provided DNA samples and provided guidance on the study. MN offered advice and guidance on the study, and revised the manuscript. All authors approved the manuscript.

The authors would like to thank Dr. JinDing Liu, Dr. Fenglong Yang, and Dr. Juan Jia from Shanxi Medical University and Dr. Xu Liu for valuable comments of this study and support. beyond human identification: Implications for development of new DNA typing systems. ELECTROPHORESIS. 1999;20:1682–96. 2. Butler JM. Genetics and genomics of core short tandem repeat loci used in human identity testing. J Forensic Sci. 2006;51:253–65. profiling provides an international reference standard for human cell lines. Proceedings of the National Academy of Sciences. 2001;98:8012–7. 4. Dirks WG, Faehnrich S, Estella IAJ, Drexler HG. Short tandem repeat DNA typing provides an international reference standard for authentication of human cell lines. ALTEX. 2005;22:103–9. 5. Matsuo Y, Nishizaki C, Drexler HG. Efficient DNA fingerprinting method for the identification of cross-culture contamination of cell lines. Hum Cell. 1999;12:149–54. relationship tests that show ambiguous STR results using autosomal SNPs as supplementary 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 cross-contamination initiative: an interactive reference database of STR profiles covering common cancer cell lines. Int J Cancer. 2010;126:303–4. structure and recent improvements towards molecular authentication of human cell lines. of Microbial Pathogens in the Little Bighorn River, Montana. International Journal of Environmental Research and Public Health. 2019;16:1097. Portable Genomics for Early Detection of Plant Viruses and Pests in Sub-Saharan Africa. Genes. 2019;10:632. 2022;23:108. quantification of short tandem repeats on signal data from nanopore sequencing. Genome Biol. 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 short tandem repeat expansions and their methylation state with nanopore sequencing. Nat STRs and SNPs using Verogen’s ForenSeq DNA Signature Prep Kit and MinION. Int J Legal Med. 2021;135:1685–93. Sequencing of a Forensic STR Multiplex Reveals Loci Suitable for Single-Contributor STR Profiling. Genes (Basel). 2020;11:381. repeat identification using a nanopore-based DNA sequencer: a pilot study. J Hum Genet. 2020;65:21–4.

Genet. 2022;56:102629. forensic autosomal STRs using the Oxford Nanopore Technologies MinION device. Forensic Sci Int 2021;37:4572–4. 31. Li H. New strategies to improve minimap2 alignment accuracy. Bioinformatics. SAMtools and BCFtools. GigaScience. 2021;10:giab008. 491 492 493 494 495 496 497 individual samples (88 DNA samples) with different amplification methods and flow cells. With ForenSeq data and PowerSeq data, we optimized the parameters of NASTRA, and conduted performance evalutation and real-time scenario tesing. b. The workflow of NASTRA: 1. Extracing aligned reads from bam file; 2. Flanking regions of each read are trimmed; 3. Reads clustering are performed to find candidate allele sequences; 4. Repeat structure of each allele is inferred by the recursive inference algorithm; 5. Genotype calling. each locus when supporting reads number (SN) is fixed. b. The genotype calling accuracy of NASTRA on each locus with different ratios of supporting reads number (SNR), when SN is 25. c. The distribution of genotyping result across downsampled data. d. The blast identity distribution of alleles between NASTRA and truth. NASTRA and STRspy on ForenSeq data for 27 STR loci. b. The deviation of genotypes between NASTRA and ground truth. c. The correlation of SN and sequencing depth (blue points represent other loci). d. The difference of sequencing depth among genotyping result. results across 46 samples on 22 autosomal STRs sequenced by R10.3 flow cells. b. Accuracy comparsion of NASTRA and STRspy on PowerSeq data for 22 STR loci. For D19S433, the allele information is missing in STRspy custom database. c. The deviation of genotypes between NASTRA and ground truth.

Figure 5. Performance on real-time scenario tesing. a. overview of genotyping results obtained from 8 DNA standard samples across 27 autosomal STRs, with variations in sequencing durations, using a R9.4.1 flow cell. b. The percentages of genotyping results achieving exact match, incomplete match or imbalance/interpretation across all 8 samples, with different sequencing durations. c. The distribution of sequencing depth in relation to different sequencing durations. 524 525