Introduction

On August 19, 2026, Merck and Moderna announced positive topline results from a Phase III clinical trial in melanoma. The study evaluated a patient-specific messenger RNA (mRNA) neoantigen therapy in combination with pembrolizumab. The therapy was previously designated mRNA-4157/V940 during development; in 2025, the World Health Organization published its recommended International Nonproprietary Name, intismeran autogene. Hereafter, it is referred to as V940 individualized neoantigen mRNA therapy.[1,2]

The Phase III study, designated INTerpath-001, evaluated V940 plus pembrolizumab in patients with completely resected, high-risk stage IIB, IIC, III, or IV cutaneous melanoma. A total of 1,137 patients were enrolled. At the prespecified interim analysis, the combination met the primary endpoint of recurrence-free survival (RFS) and the key secondary endpoint of distant metastasis-free survival (DMFS). According to the Phase III topline results announced by Merck and Moderna, this was the first positive Phase III result for an individualized neoantigen therapy and the first positive Phase III result for an mRNA-based cancer therapy.[1] As of September 1, 2026, V940 remained investigational and had not been approved for marketing; only the positive Phase III topline results had been disclosed, while the full hazard ratios, event rates, patient-subgroup results, and mature overall survival data were still pending.[1,3]

What makes this development distinctive is that part of the information used to design the therapy comes from the patient’s own tumor.

DNA sequencing of tumor tissue and peripheral blood samples is used to identify patient-specific somatic mutations, while tumor RNA sequencing is used to assess expression of the corresponding mutated genes. Computational algorithms then search these mutations for neoantigens that may be recognized by the immune system. Higher-priority candidates are combined and encoded into a patient-specific mRNA, which is subsequently formulated in a lipid nanoparticle (LNP). Different patients receive different combinations of neoantigens and different mRNA sequences, while sample analysis, computational design, mRNA production, and LNP formulation follow a shared technical framework.[2,3]

This therapeutic pathway can be summarized as four sequential transformations:

Tumor sequencing identifies patient-specific mutations → machine-learning-assisted algorithms screen and rank neoantigens → mRNA converts neoantigen information into a manufacturable patient-specific therapy → the immune system generates tumor-directed T-cell responses.

One key step lies between “detecting mutations” and “selecting neoantigens with immunotherapeutic potential.” A tumor may harbor numerous somatic mutations, but only a subset can generate mutant peptides, be presented by the patient’s own HLA molecules, and subsequently induce neoantigen-specific T-cell responses. Publicly available information indicates that machine-learning algorithms participate in the design of patient-specific therapies; during neoantigen selection, the algorithm integrates information including tumor mutations, RNA expression, amino-acid sequence conservation, and patient HLA type to screen and rank candidate neoantigens. Moderna has also disclosed that machine-learning algorithms are used to optimize the timely manufacturing and delivery of intismeran to patients, although the complete model architecture, specific parameters, and implementation details of the relevant algorithms have not been disclosed.[3,6]

1. How Tumor Mutations Become Patient-Specific Neoantigens

1.1 Neoantigens arise from tumor-specific genetic alterations

Cancer cells continuously accumulate somatic mutations during tumor initiation and evolution. Some mutations alter the amino-acid sequence of proteins, thereby generating mutant protein fragments that are absent from normal cells.

Intracellular proteins are processed and degraded into short peptides. Some mutant peptides can bind to human leukocyte antigen (HLA) molecules and be presented on the cell surface. Tumor-specific mutant peptides that can subsequently be recognized by T cells and induce an immune response represent an important source of neoantigens.[3]

The progression from a DNA mutation to a neoantigen with therapeutic value involves a continuous biological sequence:

Somatic mutation formation → mutant-gene expression → mutant-protein production → mutant-peptide formation → HLA binding and presentation → T-cell recognition and response.

Even if a mutation can be detected by DNA sequencing, its value as a therapeutic target will be limited if the corresponding gene is not expressed, the mutant peptide cannot be generated, the peptide cannot appropriately interact with the patient’s HLA molecules, or it cannot be effectively recognized by T cells.

1.2 The same type of melanoma can correspond to different neoantigen combinations

Melanomas from different patients do not have identical somatic mutation profiles.

HLA genotype introduces another layer of interindividual variation. Different HLA molecules do not bind and present exactly the same peptide repertoire. Consequently, a peptide derived from the same mutation may have favorable presentation characteristics in one patient but may not be effectively bound and presented by the relevant HLA molecules in another patient.

Patient-specific neoantigens are mainly influenced by two groups of variables:

The World Health Organization structural definition of the International Nonproprietary Name for V940 indicates that a single V940 mRNA can encode up to 34 neoantigens selected according to the patient’s tumor mutation profile and HLA type.[2]

This forms the molecular basis of “one patient, one drug”: each patient uses the same type of mRNA-LNP technology platform, but the neoantigen-encoding content carried by the mRNA varies according to the patient’s own tumor data.

2. From Tumor Sequencing to Candidate Neoantigens

2.1 What information is provided by DNA, RNA, and HLA typing?

The end-to-end V940 design and manufacturing workflow published in 2026 shows that neoantigen selection begins with the patient’s tumor tissue and peripheral blood samples and proceeds through pathological assessment, DNA and RNA extraction, next-generation sequencing (NGS), variant identification, HLA typing, and subsequent computational analysis.[3]

Different data types answer different questions:

Sequencing initially generates a relatively large set of candidate mutations. Subsequent evaluation is needed to determine whether the mutations are expressed, whether appropriate mutant peptides can be generated, and whether these peptides can match the patient’s HLA molecules.

2.2 In the TESLA validation dataset, 34 of 532 candidate peptides showed experimental immunogenicity

In 2026, Cai et al. developed the CNNeoPP neoantigen-prediction pipeline. CNNeoPP integrates tumor DNA/RNA sequencing analysis with deep-learning-based neoantigen prediction, with CNNeo serving as the deep-learning model responsible for neoantigen classification and ranking.[4]

The researchers performed independent testing using publicly available validation data established by the Tumor Neoantigen Selection Alliance (TESLA). The alliance provides a common benchmarking framework for comparing different neoantigen-prediction algorithms by using standardized patient sequencing data and experimentally validated results. Among the five patients analyzed by CNNeoPP, three had melanoma and two had non-small cell lung cancer. The dataset contained 532 experimentally tested candidate peptides derived from single-nucleotide variants, of which 34 were experimentally confirmed to be immunogenic, corresponding to 6.39%.[4,5]

These results illustrate the quantitative reduction that occurs during neoantigen screening:

Tumor mutations → candidate mutant peptides → candidates with experimentally confirmed immunogenicity

These results show that, within this independent validation subset, candidate mutant peptides are not equivalent to experimentally immunogenic neoantigens; factors including antigen expression, presentation, and T-cell recognition can all influence eventual immunogenicity.

Figure 1. From Patient Sequencing Data to Neoantigen Prediction and Experimental Validation

Panel A shows the CNNeoPP neoantigen-prediction workflow. DNA sequencing is performed on tumor tissue and peripheral blood mononuclear cells, while RNA sequencing is performed on tumor tissue. Somatic single-nucleotide variants are then identified, patient HLA type and tumor gene-expression information are obtained, mutations are converted into corresponding peptide sequences, and immunogenicity-related features are calculated. After gene-expression filtering, candidates enter the CNNeo model, which ultimately generates ranked candidate neoantigens.

Panel B presents independent validation data from five patients in the Tumor Neoantigen Selection Alliance (TESLA), including three patients with melanoma and two with non-small cell lung cancer. Across the five patients, 532 single-nucleotide-variant-derived candidate peptides were experimentally tested, of which 34 were confirmed to be immunogenic, corresponding to 6.39%.

Panel C compares the ranking performance of CNNeoPP with other neoantigen-prediction pipelines. “Top 10,” “Top 20,” and “Top 50” refer to the 10, 20, and 50 highest-priority candidates ranked by each algorithm, respectively. The height of each bar indicates how many experimentally confirmed immunogenic neoantigens were contained among those candidates. For example, among the top 10, top 20, and top 50 candidates ranked by CNNeoPP, 3, 5, and 8 experimentally confirmed neoantigens were identified, respectively.[4]

In the original figure, PBMC denotes peripheral blood mononuclear cells; SNV denotes single-nucleotide variant; and NSCLC denotes non-small cell lung cancer. The different algorithms shown in the figure were compared using the same set of candidate peptide-HLA combinations.

Image source: Cai Y et al., 2026, Figure 4. The original article is distributed under a Creative Commons Attribution (CC BY) license. The figure has not been modified.

Note: The proportion of 34 experimentally confirmed immunogenic peptides among 532 candidates was derived from the independent validation subset of five TESLA patients used by CNNeoPP. It illustrates the substantial screening challenge involved in identifying experimentally immunogenic neoantigens from a large number of single-nucleotide-variant-derived candidate peptides and should not be regarded as a fixed immunogenicity rate for V940, for all tumor types, or for all neoantigen-prediction algorithms.[4]

3. How Machine Learning Participates in Neoantigen Screening

3.1 Neoantigen screening requires simultaneous consideration of multiple biological variables

Moderna’s 2025 Annual Report explicitly disclosed that V940 uses NGS and a machine-learning-based algorithm to design mRNA according to each patient’s tumor mutations and HLA type, encoding up to 34 neoantigens while considering the potential to induce both CD8⁺ and CD4⁺ T-cell responses.[6]

The publicly available V940 workflow shows that candidate neoantigen ranking integrates multiple categories of information:[3]

① somatic mutations and their variant allele frequencies;

② RNA expression of mutated genes in the tumor;

③ amino-acid sequence conservation;

④ the patient’s HLA class I and class II genotypes;

⑤ the predicted potential to induce CD8⁺ cytotoxic T-cell and CD4⁺ helper T-cell responses.

These variables correspond to different constraints in the generation of candidate neoantigens.

Mutation information reflects what alterations have occurred in the tumor;

RNA expression reflects whether the corresponding mutation is transcriptionally expressed and therefore has the potential to generate protein;

Amino-acid-sequence-related features and HLA type influence the likelihood that a candidate mutant peptide can bind to HLA and be presented;

Immunogenicity-related features are used to further evaluate the likelihood that a candidate will be recognized by T cells and induce an immune response.

Within the neoantigen-selection step, machine learning is used to integrate these variables and prioritize large numbers of candidates.

3.2 From “Can it bind HLA?” to “Is it more likely to be immunogenic?”

Traditional computational neoantigen screening commonly uses peptide-HLA binding capacity as an important criterion, but HLA binding is only one of the requirements for an immunogenic neoantigen. Whether a candidate peptide can be generated and presented is also influenced by gene expression, protein expression, antigen processing, and TAP transport. Once a peptide-HLA complex is formed, it must also be recognized by a T-cell receptor before it can potentially induce a T-cell response.

In 2025, Lu et al. developed OmniNeo, a multi-omics neoantigen-screening pipeline. OmniNeo integrates genomic, transcriptomic, and proteomic information with computational prediction in a single workflow and further uses the deep-learning model OmniNeo-CNN to score and rank the immunogenicity of HLA class I candidate peptides.[7]

Figure 2. Stepwise Neoantigen Screening Assisted by Multi-Omics and Artificial Intelligence

The workflow begins with whole-exome or whole-genome sequencing of tumor and normal tissues together with tumor RNA sequencing. It identifies single-nucleotide variants, insertions/deletions, fusions, and variants originating from non-coding regions, followed by HLA typing, variant annotation, and extraction of candidate mutant peptides. Candidate peptides then undergo peptide-HLA binding prediction and TAP transport-efficiency prediction, followed by gene-expression filtering using tumor RNA expression. Proteomic mass-spectrometry data provide evidence for filtering at the protein-expression level. Candidates can then be further screened according to sequence similarity to experimentally validated immunogenic peptides, and HLA class I candidates are subsequently scored for immunogenicity using OmniNeo-CNN, ultimately yielding high-confidence candidate neoantigens.[7]

In the original figure, WES/WGS denote whole-exome sequencing and whole-genome sequencing, respectively; RNA-seq denotes RNA sequencing; SNV denotes single-nucleotide variant; TAP denotes the transporter associated with antigen processing, whose transport efficiency influences HLA class I antigen presentation; LC-MS/MS denotes liquid chromatography-tandem mass spectrometry, which is used in this workflow to obtain proteomic experimental information and assist in confirming protein-level expression of candidate mutant peptides; MHC denotes the major histocompatibility complex, whose corresponding molecules in humans are referred to as HLA.

Image source: Lu M et al., 2025, Figure 2. The original article is distributed under a Creative Commons Attribution (CC BY) license. The figure has not been modified.

Note: OmniNeo is an independent neoantigen-prediction study and is not the proprietary neoantigen-selection algorithm used for V940. This figure is used to illustrate how multi-omics information and deep learning can participate in the stepwise screening of candidate neoantigens; it does not represent the model architecture or specific screening workflow actually used by V940.

3.3 Publicly disclosed machine-learning information for V940 and its boundaries

Public information on V940 confirms the use of machine-learning-based algorithms and discloses major input information such as tumor mutations, RNA expression, HLA type, and sequence-related features.[3,6]

The complete model architecture, training-data composition, feature weights, and specific ranking functions of V940’s proprietary algorithm have not been disclosed.

CNNeoPP and OmniNeo are independent publicly reported studies that illustrate how sequencing, HLA information, multi-omics data, and deep learning can be used for neoantigen screening in the broader field. V940 uses Moderna’s own proprietary neoantigen-selection system.[3,4,7]

In the V940 neoantigen-selection step, the role of machine learning can be summarized as follows: prioritizing candidate neoantigens, among the large number of distinct tumor mutations in each patient, that are more likely to be presented by HLA molecules and induce T-cell responses.

4. From Computational Results to a Patient-Specific mRNA Therapy

4.1 Up to 34 neoantigens are incorporated into a single patient-specific mRNA

After neoantigen ranking, V940 selects up to 34 patient-specific candidates and combines the corresponding coding sequences into a single synthetic mRNA.[2,3]

The World Health Organization structural description of the International Nonproprietary Name for V940 indicates that its mRNA uses optimized neoantigen-encoding sequences and substitutes N1-methylpseudouridine for ordinary uridine. N1-methylpseudouridine is a modified nucleoside that can alter mRNA stability, translation, and innate immune recognition.[2]

The generation of a patient-specific therapy can be summarized as follows:

During mRNA production, T7 bacteriophage RNA polymerase is used for in vitro transcription, after which the mRNA is encapsulated in LNPs to generate a patient-specific batch.[3]

4.2 “One patient, one drug” depends on both front-end data quality and back-end manufacturing capacity

Production of an individualized therapy begins with the patient’s samples. Problems at any stage—including sample quality, DNA/RNA extraction, sequencing, neoantigen identification, or mRNA manufacturing—may affect the generation of the final therapy.

A manufacturing study published in 2026 showed that, among patients in whom neoantigen selection was successfully completed, the subsequent V940 manufacturing success rate exceeded 99%. Under clinical-trial conditions, approximately six weeks were required from submission of qualified tumor tissue and peripheral blood samples to delivery of the patient-specific formulation to the clinical site.[3]

This model requires five categories of information to remain accurately linked:

Patient identity—tumor sample—sequencing data—mRNA sequence—drug batch.

The programmable nature of mRNA enables different patients’ antigen sequences to enter the same type of manufacturing platform. The ability to scale individualized therapy simultaneously depends on sequencing speed, algorithmic computation, mRNA production, quality control, and logistics.

5. How Neoantigen mRNA Is Converted into Antitumor Immunity

5.1 mRNA converts digital sequence information into antigens recognizable by the immune system

After V940 enters the body, cells use the genetic information in the mRNA to synthesize the corresponding neoantigen protein fragments. These antigens are subsequently processed intracellularly to form peptides that can be presented by HLA molecules.

HLA class I molecules mainly present antigens to CD8⁺ cytotoxic T cells, whereas HLA class II molecules mainly participate in CD4⁺ helper T-cell responses. V940 candidate selection considers both classes of antigen presentation and T-cell response.[3,6]

This process completes another layer of information conversion in therapeutic design:

Patient-specific sequences in a computer → neoantigens expressed inside cells → tumor markers recognizable by T cells.

5.2 A melanoma animal model shows that neoantigen mRNA-LNP can induce T-cell responses and delay tumor growth

In 2022, Sittplangkoon et al. used the B16F10 mouse melanoma model to investigate the immune and antitumor effects of neoantigen mRNA-LNP. Figure 8 of the original study used luciferase-labeled B16F10-Luc2 tumor cells to establish a subcutaneous melanoma model in mice.

The study selected two somatic mutations from the Pbk and Actn4 genes as neoantigen sources. Pbk encodes PDZ-binding kinase, whereas Actn4 encodes α-actinin-4. The coding sequences for the two mutant neoantigens were combined into a single mRNA construct and delivered using lipid nanoparticles.[8]

The study compared neoantigen mRNA-LNP containing unmodified uridine with mRNA-LNP in which all uridine was replaced by N1-methylpseudouridine (m1Ψ). Neoantigen mRNA-LNP containing unmodified uridine enhanced antigen-specific T-cell responses, increased activation-related indicators in intratumoral T cells and type 1 conventional dendritic cells, and markedly delayed tumor growth. In the survival analysis, the authors reported that approximately one-third of mice treated with neoantigen mRNA-LNP (Neo-LNP) survived to day 35, whereas all mice in the control groups reached the experimental endpoint before day 29.[8]

Figure 3. Immune and Antitumor Effects of Neoantigen mRNA-LNP in a Mouse Melanoma Model

Panel A shows the B16F10-Luc2 mouse melanoma model and dosing schedule; panel B measures the proportion of granzyme B-positive cells among splenic CD8⁺ T cells; panel C compares tumor volumes among different treatment groups; panel D measures the proportion of CD69-positive cells among tumor-infiltrating CD4⁺ and CD8⁺ T cells; panel E measures expression of the co-stimulatory molecule CD40 in tumor-infiltrating type 1 conventional dendritic cells; and panel F compares animal survival among the different treatment groups.[8]

In the original figure, GzmB denotes granzyme B, an important molecule involved in the effector function of cytotoxic T cells; CD69 is a T-cell activation-associated marker; cDC1 denotes type 1 conventional dendritic cells, which play important roles in tumor-antigen cross-presentation and the priming of CD8⁺ T cells; CD40 is a co-stimulatory molecule on the surface of dendritic cells and is associated with dendritic-cell activation and initiation of T-cell immunity; m1Ψ denotes N1-methylpseudouridine.

Image source: Sittplangkoon C et al., 2022, Figure 8. The original article is distributed under a Creative Commons Attribution (CC BY) license. The figure has not been modified.

Note: This study compared unmodified-uridine mRNA with N1-methylpseudouridine-modified mRNA and used its own lipid nanoparticle system. Its mRNA construct design and lipid nanoparticle system differ from those of V940. The figure is used to demonstrate the experimental relationship between “neoantigen encoding—mRNA delivery—T-cell immune response—tumor control” in an animal model and does not provide direct evidence for the specific formulation or clinical efficacy of V940.

5.3 Pembrolizumab reduces PD-1-mediated inhibition of T cells

V940 and pembrolizumab serve different therapeutic functions.

PD-1 is an inhibitory immune checkpoint expressed on T cells and other immune cells. Binding of PD-1 to PD-L1 or PD-L2 generates inhibitory signaling that reduces T-cell activity. Pembrolizumab blocks PD-1 and thereby reduces this inhibitory signal, creating a more favorable immune environment in which neoantigen-specific T cells can sustain antitumor activity.

The combination therefore forms a sequential relationship:

Patient-specific neoantigens provide recognition targets → neoantigen-specific T cells are generated or expanded → PD-1 blockade reduces inhibition of T cells.

5.4 T cells targeting V940-encoded neoantigens have been detected in patients

At the 2026 American Society of Clinical Oncology Annual Meeting, results from studies of T-cell responses after V940 treatment were reported.

Among three patients treated with V940 plus pembrolizumab, investigators used T-cell receptor (TCR) sequencing and functional experiments to identify 20 T-cell receptor clones that expanded after treatment. These clones corresponded to 11 neoantigens actually encoded by V940. Some responses remained detectable more than 100 days after the final V940 administration.[9]

Functional experiments in an additional seven patients showed that neoantigen-reactive CD8⁺ T cells produced interferon-γ and tumor necrosis factor-α following stimulation with the corresponding mutant peptides and exhibited degranulation associated with cytotoxic activity.[9]

Note: These findings are derived from mechanistic analyses in a small number of patients. They provide direct evidence that V940-encoded neoantigens can induce corresponding T-cell responses but cannot be used to estimate the neoantigen-specific T-cell response rate across the entire treated population.[9]

6. From Patient-Specific Design to Phase III Clinical Benefit

6.1 The Phase IIb study established a clinical efficacy signal

V940 plus pembrolizumab initially demonstrated a clinical efficacy signal in a randomized Phase IIb study.

The study, designated KEYNOTE-942, evaluated V940 plus pembrolizumab in patients with completely resected, high-risk stage IIIB-IV cutaneous melanoma. A total of 157 patients were enrolled, including 107 who received V940 plus pembrolizumab and 50 who received pembrolizumab alone.[10]

Initial results published in 2024 showed 24 recurrences or deaths in the combination group and 20 in the monotherapy group. The 18-month RFS rates were approximately 79% and 62%, respectively, with an RFS hazard ratio of 0.561 (95% CI: 0.309-1.017; two-sided P=0.053 as reported in the paper). The Phase IIb study used a prespecified one-sided statistical design according to the protocol, with a one-sided P value of 0.0266, thereby meeting the study’s prespecified statistical criterion.[10]

The 5-year follow-up reported in 2026 was a descriptive analysis for which no formal alpha level was allocated. Results for the combination group relative to pembrolizumab monotherapy were as follows:[11]

① the RFS hazard ratio was 0.510 (95% CI: 0.294-0.887), corresponding to an approximately 49% relative reduction in the instantaneous hazard of recurrence or death;

② the DMFS hazard ratio was 0.411 (95% CI: 0.200-0.843), corresponding to an approximately 59% relative reduction in the instantaneous hazard of distant metastasis or death;

③ the OS hazard ratio was 0.471 (95% CI: 0.165-1.345), showing a currently favorable trend, although the confidence interval still crossed 1 and an OS benefit cannot yet be confirmed.

6.2 The Phase III study expanded the validation population to 1,137 patients

The INTerpath-001 study reported in 2026 expanded the study population to 1,137 patients and included patients with completely resected stage IIB, IIC, III, or IV cutaneous melanoma.

At the prespecified interim analysis, both the primary endpoint of RFS and the key secondary endpoint of DMFS were met. The companies reported that both improvements were statistically significant and clinically meaningful, that the safety profile was generally consistent with previous studies, and that no new safety signals were identified.[1]

Five sequential levels of evidence have now been established:

From a patient’s tumor specimen to an independently manufactured mRNA therapeutic formulation, the process involves sequential steps including tumor-data acquisition, neoantigen screening, sequence design, and drug manufacturing; after administration, an immune response against the encoded neoantigens must then be generated.

6.3 Neoantigen prediction still needs to improve identification of “effective candidates”

Neoantigen prediction continues to involve substantial screening challenges. In the TESLA independent validation subset described above, experimentally confirmed immunogenic candidates accounted for only a small proportion of the total. This finding shows that candidate mutant peptides derived from somatic mutations cannot automatically be assumed to be immunogenic, and prediction needs to integrate additional data that more directly reflect antigen presentation and T-cell recognition.[4,5]

Prediction accuracy may be further improved by integrating data that more directly reflect antigen presentation and T-cell responses:

Computational neoantigen prediction is evolving from screening centered on peptide-HLA binding toward approaches that comprehensively incorporate antigen expression, antigen presentation, T-cell recognition, and functional effects.[4,5,7]

6.4 Individualized manufacturing turnaround time affects treatment accessibility and applicable clinical settings

In the clinical trials described above, the turnaround time from submission of qualified tumor tissue and peripheral blood samples to delivery of the patient-specific formulation to the clinical center was approximately six weeks.[3]

This period includes:

Sample processing → sequencing → computational neoantigen screening → mRNA sequence design → mRNA production → LNP formulation → quality release → transportation.

Adjuvant therapy after surgery can provide a certain manufacturing time window. For rapidly progressing advanced tumors, the manufacturing period may become a more prominent clinical limitation.

Shortening the turnaround time for patient-specific therapies, improving the success rate of front-end sequencing and design, and expanding the capacity for parallel production of patient-specific batches, together with improved neoantigen-prediction accuracy, will influence the accessibility of this therapeutic model.

6.5 Extension of the melanoma results to other tumor types still requires independent clinical validation

V940 is currently also being studied in non-small cell lung cancer, renal cell carcinoma, bladder cancer, and other melanoma treatment settings.[1,6]

Different tumor types vary in somatic mutation burden, neoantigen quality, HLA presentation capacity, intratumoral T-cell infiltration, and the immunosuppressive microenvironment.

The Phase III results in 2026 further demonstrate that highly heterogeneous tumor mutations among patients can be converted through sequencing and computational screening into patient-specific neoantigen-encoding information and subsequently into a manufacturable and administrable mRNA therapeutic formulation. INTerpath-001 enrolled 1,137 patients, and the positive topline results showed that both the RFS primary endpoint and the DMFS key secondary endpoint were met.[1]

At the therapeutic-design level, a key task of machine learning is to screen and rank, from large numbers of patient-specific mutations, neoantigens that are more likely to have immunotherapeutic value. Moderna has also disclosed that machine learning is used to optimize the timely manufacturing and delivery of intismeran.[6]

This model is gradually transforming interpatient tumor heterogeneity from an obstacle in drug development into an input for drug design itself.

Note: The products listed below are intended for research related to mRNA nucleosides, in vitro transcription, and lipid-mediated delivery. Public information on V940 indicates that its mRNA uses N1-methylpseudouridine in place of uridine and is delivered using lipid nanoparticles (LNPs), but the specific LNP formulation has not been disclosed. Therefore, the lipids listed in Tables 2 and 3 should be regarded as representative research materials in related fields and do not represent the actual formulation components of V940.[2,3,6]

Table 1. RNA Nucleosides, Modified Nucleosides, and In Vitro Transcription Materials

Table 2. Ionizable and Cationic Lipids for Nucleic Acid Delivery Research

Table 3. Lipids for Studying Lipid Nanoparticle Structure, Membrane Behavior, and Surface Stabilization

Note: The products listed above are representative Aladdin products for related research. Specific applications should be determined according to product specifications, batch-specific Certificates of Analysis (COAs), and the intended reaction or evaluation system. Additional product specifications, grades, and COA information can be searched on the Aladdin website using the product name, CAS number, or catalog number.

References

[1] Merck & Co., Inc.; Moderna, Inc. Merck and Moderna Announce Phase 3 INTerpath-001 Trial of Intismeran Autogene Plus KEYTRUDA Met Endpoints of Recurrence-Free Survival and Distant Metastasis-Free Survival in Patients With Completely Resected Stage IIB-IV Melanoma. 2026-08-19.

[2] World Health Organization. Recommended International Nonproprietary Names (INN): List 93. WHO Drug Information. 2025;39(1):192–193.

[3] Ojalvo LS, Posadas T, Chattopadhyay N, et al. Intismeran Autogene Therapy: An End-to-End Pathway From Tumor Tissue to Individualized Neoantigen Treatment for Patients With Cancer. JCO Oncology Advances. 2026;3. DOI: 10.1200/OA-25-00206.

[4] Cai Y, Chen R, Song M, et al. CNNeoPP: a large language model-enhanced deep learning pipeline for personalized neoantigen prediction and liquid biopsy applications. Frontiers in Immunology. 2026;17:1722117. DOI: 10.3389/fimmu.2026.1722117.

[5] Wells DK, van Buuren MM, Dang KK, et al. Key Parameters of Tumor Epitope Immunogenicity Revealed Through a Consortium Approach Improve Neoantigen Prediction. Cell. 2020;183:818–834.e13. DOI: 10.1016/j.cell.2020.09.015.

[6] Moderna, Inc. 2025 Annual Report. Intismeran autogene development section. 2026.

[7] Lu M, Liu Y, Xu L, et al. OmniNeo: a multi-omics pipeline incorporating proteomics and AI selection for neoantigen optimization in tumor immunotherapy. Frontiers in Immunology. 2025;16:1727642. DOI: 10.3389/fimmu.2025.1727642.

[8] Sittplangkoon C, Alameh MG, Weissman D, et al. mRNA vaccine with unmodified uridine induces robust type I interferon-dependent anti-tumor immunity in a melanoma model. Frontiers in Immunology. 2022;13:983000. DOI: 10.3389/fimmu.2022.983000.

[9] Sullivan RJ, Carlino MS, Khattak MA, et al. Intismeran autogene to induce de novo neoantigen-specific T cells as adjuvant therapy in melanoma. Journal of Clinical Oncology. 2026;44(16_suppl):9564. DOI: 10.1200/JCO.2026.44.16_suppl.9564.

[10] Weber JS, Carlino MS, Khattak A, et al. Individualised neoantigen therapy mRNA-4157 (V940) plus pembrolizumab versus pembrolizumab monotherapy in resected melanoma (KEYNOTE-942): a randomised, phase 2b study. Lancet. 2024;403:632–644. DOI: 10.1016/S0140-6736(23)02268-7.

[11] Khattak A, Carlino MS, Meniawy T, et al. Intismeran Autogene Plus Pembrolizumab Versus Pembrolizumab Alone in High-Risk Resected Melanoma: 5-Year Update of the Randomized Phase IIb KEYNOTE-942 Study. Journal of Clinical Oncology. 2026. DOI: 10.1200/JCO-26-00835.

For more related articles, see below:

Culture Systems, Phenotypic Characterization, and Translational Research Applications of Tumor Organoids

HMGB1-Mediated Damage-Associated Inflammatory Signaling and Tumor Immune Regulation

Structural Features, Engineering Design, and Applications of Nanobodies in Tumor Diagnosis and Therapy

Role of the ANGPT2/TIE2 Signaling Axis in Tumor Angiogenesis and Anti-Angiogenic Therapy

Role of the HER3 Signaling Network in Tumor Progression, Therapeutic Resistance, and Targeted Intervention

How NKG2A and NKG2D Jointly Regulate Tumor Immune Killing

How Does Dynamic ECM Remodeling in the Tumor Microenvironment Regulate Tumor Immune Responses?

Nucleotide Synthesis in Cancer Cells