Establishment of long-term response prediction method against immune checkpoint inhibitors using machine learning by artificial intelligence (AI)

Preferred Medicine, Inc

September 21st, 2022


Distinguishing cancer patients who are expected to be highly effective before administration

Yifan Zhang of Preferred Networks, Nobuyuki Ota of Preferred Medicine CEO (Preferred Medicine an early detection miRNA based liquid biopsy company based in the Bay Area, CA focused on early detection of breast cancer as its initial target) in collaboration with Yu Fujita, Lecturer at Jikei University Tokyo Jikei University School of Medicine, Exosome Drug Discovery Research Course, Tei Goto, Department of Respiratory Medicine, National Cancer Center Hospital, have constructed an algorithm that enables the prediction of patients(exceptional-responders or super-responders) whose effects will continue over the long term for immune checkpoint inhibitors (ICIs) by machine learning techniques based on artificial intelligence (AI) using genetic information in body fluids such as microRNA Note1 .

The efficacy of ICIs such as anti-PD-1/PD-L1/CTLA4 antibodies has recently been reported in the treatment of various carcinomas including lung cancer. On the other hand, ICIs are expensive and can cause fatal immune-related side effects. Furthermore, the therapeutic effect of ICIs varies depending on the patient, and there are patients who have no effect at all(Resistance) and patients who can obtain a long-term response or complete remission (Exceptional-responder). Predictive testing is not currently sufficient. As shown in this study, if this diagnostic method with high accuracy before treatment is proved to be useful, it will be possible to select appropriate drugs for patients who are expected to be effective, and it will be possible to reduce problems caused by side effects.

The research results of this industry-academia collaboration will be published in the online version of the international scientific journal "Lung Cancer" (link inside) (published on September 13th 2022, Japan time). See link below:


・Pre-treatment prediction of exceptional-responders of ICIs can select suitable patients who will receive the maximum benefit from treatment.

・Therefore, we collected pretreatment serum and various clinical information in 213 advanced-stage non-small cell lung cancer (NSCLC) patients treated with the anti-PD-1 antibody nivolumab monotherapy.

・Based on 45 types of serum microRNA expression and 3 clinical information, we constructed a machine learning model optimized to identify exceptional-responders .

・This algorithm significantly outperformed conventional companion diagnostic reagents Note2 in identifying exceptional-responders.

Research details

Conceptual Diagram of Research
Conceptual Diagram of Research


1. Background

Currently, cancer immunotherapy is making remarkable progress, and ICIs such as anti-PD-1/PD-L1 antibody and anti-CTLA-4 antibody are effective against various cancers such as malignant melanoma, renal cancer, and lung cancer. Strong and durable tumor regression effects, including complete remission, have been observed in some patients.

Recently, treatment options have diversified, such as combination with cytocidal anticancer drugs and combination with ICIs, but the response rate of ICI monotherapy is about 20-40% in all reports. There is also a population of patients who do not respond at all. Until now, tumor PD-L1 immunostaining, which predicts response, has been approved as a companion diagnostic agent, but response may be observed even in negative cases, and a more accurate method for predicting response to treatment has been required. In particular, pretreatment prediction of exceptional-responders in ICIs allows selection of suitable patients who can receive the maximum benefit from treatment, but until now only a small population of long-term responders has been identified. Few targeted predictive biomarker studies have been performed.

MicroRNAs are short (20-25 nucleotides)RNAs (non-coding RNAs) that do not produce proteins. These short RNAs are known to regulate the expression of other genes, and have been found to play important roles in life phenomena and diseases from previous studies. Furthermore, microRNAs exist stably in body fluids by being encapsulated in extracellular vesicle granules ( including exosomes ), etc., and are involved in cancer metastasis and malignant transformation mechanisms as communication tools between cells. has been reported. We previously reported that microRNAs are involved in regulation of the expression of the immune checkpoint molecule PD-L1 (Fujita Y, et al. MolTher , 23(4):717-27, 2015). , The research team believes that a highly accurate prediction model can be constructed by building a machine learning model using artificial intelligence, with various clinical information attached, centering on the analysis of microRNA in blood., conducted a study.

2. Research Methods and Results

(i) Research Method

Pretreatment serum and various clinical information were collected in NSCLC’s patients treated within volumab monotherapy at the National Cancer Center Hospital . Patients were divided into three groups based on treatment response and progression -free survival (PFS) . Exceptional-responders were defined as either complete remission (CR) or partial response (PR) and PFS >20months ( 27 cases in total ) . Non-response patients (Resistance) , on the other hand, had stable disease (SD) or progressive disease (PD) , and had PFS within20 months ( total of 161 patients ) . Patients not included in long- term responders and non-responders were grouped into 3 groups ( 213 total ) , Other (25 total ) . Non-responders were further randomized into 6 non-responders (26or 27 each), and each split was combined with long-term responders and the Others group for a balanced data set .

Using serum from these patients before treatment with nivolumab, RNA was collected and RNA- seq analysis was performed using a next-generation sequencer to evaluate the expression of microRNAs. In addition to these, the results of tumor PD-L1 immunostaining ( strongly positive ( ≥50 % ) , positive ( 1-50 %), negative ( < 1% ) and unexplored (NE)) , tumor histology ( adenocarcinoma, squamous ) Patient clinical information such as epithelial cancer, others),gender ( male, female ) , smoking history ( yes, no ) , Draver gene mutation ( yes, no), age (75 years old or younger, 75 years old or older) Used. The purpose of machine learning is to classify the three groups of Exceptional-responders , Resistance, and Others , and we used the Random Forest algorithm to handle these diverse data.

(ii) Outcomes

We built a machine learning model optimized for identifying exceptional-responders by three-group classification, and built a panel consisting of 45 types of microRNAs and clinical information from three fields . The machine learning model based on this panel should achieve a sensitivity of 0.81-0.89 (median 0.85) and an accuracy of 0.52-0.71 (median 0.59) in 3-group classification by 5 -fold cross-validation for [A2] all 6 datasets constructed.(Exceptional-responder vs Other , AUC=0.746; Exceptional-responder vs Resistance, AUC=0.830) . Tumor PD-L1 immunostaining using a conventional companion diagnostic has a sensitivity of 0.44-0.44 and an accuracy of0.55-0.67 (median value 0.62 ) . It was shown to be a much higher predictive model. In addition, several previous studies have reported the relationship between the efficacy of ICIs and microRNA panels in NSCLC patients , but our microRNA panel predicted the highest number of exceptional-responders. showed diagnostic ability. For these 45 microRNAs , factors such as miR-21-5p , miR -429 , miR-100-5p, miR-141-3p , miR-200c-3p , miR-200a-3p are mediated in tumor cells. Reportedto regulate the PD-1/PD-L1 pathway, confidence that our predictive model with unique microRNAs reflects changes in tumor-host immune interactions and shows superior reactivity It was an endorsement that it could be a tool that can.

3. The Next Deployment

Research group will collaborate with pharmaceutical companies to evaluate the usefulness of the algorithm established by the results of this study in various ICIs other than nivolumab, and will conduct verification in prospective studies. In addition , we will further investigate whether non-invasive treatment effect prediction models using microRNA can be effectively used for treatment effect monitoring and side effect prediction during the course of treatment.

4. Footnotes, Glossary

Note 1 MicroRNA: A micro RNA consisting of 20 to 25 nucleotides encoded on the genome of a gene. It is a functional nucleic acid and plays an important role in various phenomena in vivo by regulating the expression of other genes. The name of each microRNA is written as "miR".

Note 2 Companion diagnostics: Preliminary tests to determine whether a therapeutic agent is effective for a patient before treatment. It is used to advance personalized medicine and is distinguished from routine laboratory testing. PD-L1 immunostaining corresponds to ICIs.

5. Contact Point for Inquiries

Preferred Medicine Inc. :

Stay Informed

Sign up to be the first to hear any updates

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Please try again.
We use cookies to provide you with a greater online experience. By using our sites, you agree to our use of cookies. See our cookie policy.