Scipher: Decoding Drug Response From Protein Interactions

Scipher aims to improve drug response predictions by incorporating protein interaction networks in its gene expression analysis algorithms.

To improve the accuracy of therapeutic response predictions, Scipher has incorporated protein-protein interaction map into its machine learning-based analyses of patient gene expression data to spotlight pathologically relevant markers.

According to Scipher Medicine Corp. CEO Alif Saleh, 90% of patients prescribed the world’s 10 top-selling drugs do not respond to therapy. The company described the lack of efficacy to the irrelevance of the drug target in non-responding patients.

Founded in 2014 by Northeastern University’s Albert-László Barabási and Brigham and Women’s Hospital’s Joseph Loscalzo, Scipher aims to reduce the rate of treatment failure as well as identify efficacious mechanisms of action for non-responders using indication-specific tests derived from its PrismIX machine-learning platform, starting with autoimmune diseases.

The company’s founding IP was developed by Barabási, a professor of network science at Northeastern’s Khoury College of Computer Sciences, and covers the protein network map and the platform’s core algorithms. Loscalzo is chair of the Department of Medicine and physician-in-chief at Brigham and Women’s.

Other companies with machine-learning platforms that analyze gene expression data to predict drug responses or identify drug targets include WuXi NextCode Genomics Inc., Berg LLC and Verge Genomics Inc.

While declining to comment on specific companies, Saleh said most machine-learning algorithms, which analyze sequencing data to identify correlations, identify markers that often fail to predict drug response in independent patient cohorts. The company said many of the validation failures occur because researchers typically focus on differentially expressed genes that prior studies have linked with drug response, creating a bias.

By contrast, Scipher said its platform is unbiased because its algorithms identify the portion of the protein interaction network, about 100-150 proteins, that defines a disease’s biology. For any given indication and drug, PrismIX does this by analyzing sets of patient whole blood gene expression data, typically containing about 25,000 data points, in the context of the protein interaction map to identify markers that best predict non-response. Correspondingly, PrismIX identifies the proteins that can be targeted in a patient to elicit a therapeutic response. Scipher’s platform allows it to “take those 25,000 expression data points and reduce it down to the 100 that really make up the disease biology by superimposing them on our protein network,” Saleh said.

After consulting with payers and clinicians to identify indication-drug pairs for which Scipher’s technology could deliver the most benefit in terms of cost and patient care, the company develops diagnostic tests based on PrismIX’s output. Its first products are PrismRA and PrismUC, rheumatoid arthritis and ulcerative colitis tests to predict response to anti-tumor necrosis factor (TNF) drugs. Scipher is collecting data to enable the tests to also predict response to Janus kinase (JAK)- and interleukin-targeted therapies.

Additionally, Scipher is collaborating with Beth Israel Deaconess Medical Center to enable drug response predictions in inflammatory bowel disease patients. While it has multiple academic partners for biomarker discovery, the company does not plan to develop diagnostic tests in partnerships with drug companies because it believes doing so could bias diagnostic development in favor of partners’ drugs.

The company has completed observational trials for PrismRA and PrismUC and plans to publish the data in peer reviewed journals. Saleh said Scipher has achieved a 90% accuracy across the two diagnostics but did not disclose details. It is testing the ulcerative colitis and RA tests in international interventional trials.

Although the company is now focused on anti-inflammatories, Saleh said it plans to develop tests for other top-selling drugs. Scipher has raised $10 million to date, including $8.5 million in series A funds, and is raising money for its next financing round.

Northeastern has filed patent applications covering the algorithms and treatment-related use of the diagnostics.


Scipher Medicine Corp.
Waltham, Mass.

Technology: Machine-learning platform combining a protein-protein interaction map with patient gene expression to predict drug response
Disease focus: Autoimmune
Clinical status: Preclinical
Founded: 2014 by Albert-László Barabási, PhD and Joseph Loscalzo, MD
University collaborators: Northeastern University, Brigham & Women’s Hospital/Harvard Medical School and Beth Israel Deaconess Medical Center
Corporate Partners: N/A
Number of employees: 15
Funds raised: $10 million
Investors: Khosla Ventures
CEO: Alif Saleh
Patents: None

Companies and Institutions Mentioned
Berg LLC, Framingham, Mass.
Beth Israel Deaconess Medical Center, Boston, Mass.
Brigham and Women’s Hospital, Boston, Mass.
Northeastern University, Boston, Mass.
Verge Genomics Inc. San Francisco, Calif.
WuXi NextCode Genomics Inc., Cambridge, Mass.