Identification of high-risk amino-acid substitutions in hematopoietic cell transplantation: a challenging task

Abstract

Allogeneic hematopoietic cell transplantation (HCT) offers the potential to cure hematologic malignancies. Absent an HLA matched donor, HLA mismatched unrelated donors may be used although risks of graft-versus-host disease (GvHD) and treatment related mortality (TRM) are higher. Identification and avoidance of amino acid substitutions and position types (AASPT) conferring higher risks of TRM and GvHD would potentially improve the success of transplantation from single HLA mismatched unrelated donors. Using random forest and logistic regression analyses, we identified 19 AASPT associated with greater risks for at least one adverse transplant outcome: grade III-IV acute GvHD, TRM, lower disease free survival or worse overall survival relative to HLA matched unrelated donors and to other AASPT. When tested in an independent validation cohort of 3,530 patients, none of the AASPT from the training set were validated as high-risk, however. Review of the literature shows that failure to validate original observations is the rule and not the exception in immunobiology and emphasizes the importance of independent validation prior to clinical application. Our current data do not support avoiding any specific class I AASPT for unrelated donors. Additional studies should be performed to fully understand the role of AASPT in HCT outcomes.

Artificial Intelligence approaches in hematopoietic cell transplant: A review of the current status and future directions

“…Different studies investigated the possible use of AI methods and tools to tackle this challenge. Marino et al [ 17 ] identified 19 amino acid substitutions related to at least one bad outcome following HCT using random forest and logistical regression methods. These included overall survival, treatment-related mortality, incidence of graft-versus-host disease (GVHD), etc.…”

A Systematic Review of Machine Learning Techniques in Hematopoietic Stem Cell Transplantation (HSCT)

“…ADT technique was used in four of the reviewed studies where it was used primarily to predict survival/death and relapse post-HSCT [17,24,25,27]. ADT outperformed RF model in one study [16,20,24,26,35,36,40]. Finally, logistic regression was used in six of the primary studies as a comparator technique, but it did not outperform any of the other ML techniques.…”

“…There are some studies [19,20] that applied ML techniques to identify specific genes responsible for relapse in patients undergoing HSCT. Ritari et al [19] performed a genomic-wide-sequencing of active immunoregulatory regions, whole exome, and MHC regions on 151 HSCT recipients with HLA matched donors.…”

“…The results showed that germline genetic variations were highly associated with relapse. Marino et al [20] identified 19 amino acid substitutions associated with risk of adverse outcomes following HSCT using ML techniques. The outcomes included incidence of grade III-IV acute GVHD, disease-free survival, and overall survival.…”

Machine Learning Applications in the Diagnosis of Benign and Malignant Hematological Diseases

“…The majority of ML and DL studies in HCT are for prognostication, but multiple studies have tried to tackle other problems, including pre-transplant patient’s selection. Donor/recipient and transplantation characteristics were used to create algorithms to improve pre-transplant match results [ 43 , 44 ]. Nonetheless, AI/ML is complex and dependent on multiple factors; thus, some algorithms had suboptimal accuracy or failed validation [ 43 , 45 ].…”