e20506 Background: Differentiation of multiple primary lung cancer (MPLC) from intrapulmonary metastasis (IPM) is critical to determine clinical stage. Although clinicopathological features could provide certain evidences, it’s still challenging to identify the tumor malignancy accurately. In General, standard histopathologic approach is adequate in most cases, but has notable limitations in the recognition of IPMs. Herein, we propose an integrated molecular algorithm to facilitate MPLCs and IPMs diagnosis in the clinical practice. Methods: 40 Chinese patients with lung adenocarcinomas were enrolled in the study, 84 tumor samples were collected for next-generation sequencing. Somatic alterations with variant allele fraction (VAF) ≥1% were taken into account for molecular algorithm. A genomic database of 2,471 Chinese lung adenocarcinomas (LUAD) was used to calculate odds of coincidental occurrence, prevalence of individual mutation prevalence. Tumor relatedness diagnosed by histopathologic assessment was contrasted with comparative genomic profiling by subsequent NGS. Moreover, the performance of molecular algorithm prediction was evaluated as well. Results: Firstly, we compared the performance of comprehensive next-generation sequencing (NGS) with standard histopathologic approaches for distinguishing NSCLC subtypes in clinical practice. The genomic profiling was described as following: EGFR alterations occurred more frequently in MPLCs compared to IPMs (77.1% vs 50.0%, P<0.05). Further analysis showed that TP53 alterations occurred less frequently in MPLCs compared to large Chinese cohort (22.9% vs 51.0%, P<0.05). TP53 alterations occurred less frequently in MPLCs compared to large Chinese cohort (P<0.05). The classifications based on the three different methodologies mentioned above were compared. Molecular algorithm prediction was concordant with NGS in 21 cases (52.5%), particularly in the prediction of MPLC. Retrospective review highlighted several histologic challenges, including morphologic progression in some IPMs. For the five undetermined cases, two showed differences in architectural patterns, and remained cases have nodules presented as adenocarcinoma in situ, or minimally invasive adenocarcinoma. Of 28 MPLC cases defined by NGS, 25 cases had unique somatic mutations per pair Based on calculation from the prevalence of EGFR L858R mutation (27%) in large Chinese cohort, the odds of coincidental occurrence of the mutation in two unrelated tumors was 7.3%. Taking together, EGFR alterations occurred more frequently in MPLCs compared to IPMs (77.1% vs 50.0%, P<0.05). Molecular algorithm prediction was concordant with NGS in 21 cases (52.5%). Conclusions: Our results support broad panel NGS to assist differential diagnosis to assist approach in clinical practice. It is necessary to conduct large clinical study to establish comprehensive algorithm models to assist diagnosis and predict clinical outcome.
Using Deep Learning for Gene Detection and Classification in Raw Nanopore Signals
