Lunit said it will present five poster presentations featuring its AI-biomarker platform, Lunit SCOPE IO,  during the 2023 American Association for Cancer Research (AACR) annual meeting in Orlando, Florida, from April 14-19.

One of the studies to be presented evaluates a deep learning-based ensemble model designed to predict the KRAS G12C mutation. This mutation is the most common among KRAS gene mutations and is found in approximately 25 percent of non-small cell lung cancer  (NSCLC) patients.

Lunit said researchers employed an AI model, developed using samples from The Cancer Genome Atlas LUAD and LUSC (TCGA-Lung), to conduct a novel approach aimed at improving the performance of KRAS G12C prediction.

As a result, the prediction model showed improved accuracy when compared to previously reported studies on predicting KRAS mutations.

The KRAS G12C mutation prediction model, based on Lunit SCOPE, demonstrated a high predictive power with an AUC (Area Under the Curve) of 0.787, indicating the accuracy of the AI algorithm. Furthermore, in validation with independent external data, the model showed an AUC of 0.745.

AUROC (Area Under the Curve Operating Characteristics Curve) is an indicator that evaluates the performance of an AI model expressed as a value between 0 and 1. A higher AUROC indicates better performance. 

“As the KRAS G12C mutation has become targetable in NSCLC, tissue-based KRAS mutation tests are now an essential practice for treatment decisions,” Lunit's Chief Medical Officer Ock Chan-Young said. “By combining simple H&E analysis model with Lunit SCOPE IO's in-depth predictive features, our novel approach showed significant improvement in prediction.”

Lunit believes that such a model will be able to provide predictive results before applying molecular testing, which is relatively time-consuming and expensive, and may help enable rapid treatment decisions, Ock added.

Another study Lunit plans to present during the conference demonstrates the effectiveness of universal immunohistochemistry (UIHC), an AI-powered image analyzer, in detecting and quantifying untrained new targets of interest expressed in multiple cancer types.

The company trained the AI model on the dataset of PD-L1 and HER2-stained lung, bladder, and breast cancer slides, and evaluated its performance on the hold-out IHC dataset of untrained targets and cancer types.

Compared to the AI models trained with a single IHC and cancer type, Lunit stressed that the UIHC showed superior performance for new IHC and cancer types.

“Researchers concluded that UIHC model will be a useful tool for the future clinical research targeting novel tumor-associated antigens,” the company said.

Lunit will also present additional studies during the AACR 2023 conference, including demonstrating the effectiveness of Lunit SCOPE IO as a diagnostic aid in the treatment of various cancer types, analyzing the distribution of TILs and associated genomic signatures based on proximity to the tumor-stromal border (TSB) in the TCGA pan-carcinoma dataset, and using Lunit SCOPE IO in the TCGA ovarian cancer dataset to demonstrate the enrichment of inflammatory immune and transcriptomic traits in the Inflamed IP classified by the AI solution.

“We are excited to bring new research using Lunit SCOPE in more cancer types and treatment settings,” Lunit CEO Suh Beom-seok said. “Lunit will continue to enable novel academic research and innovative product development to provide the most appropriate treatment for cancer patients.