Immunomodulating Agents

Our text mining is based on the immunosuppression-related abstracts from PubMed. Self-developed ontology-based bio-entity recognizer was used to perform bio-entity recognition and extraction from these abstracts for human immunosuppression gene candidates. Our recognition tool has the precision, recall and F-measure of 0.81, 0.88 and 0.85 against the CRAFT corpus for gene/protein recognition based on Protein Ontology (PR), which are comparable to current state-of-the-art biomedical annotation systems like BeCAS (13 (link)).
Three steps were taken to compile the list of human immunosuppression genes with their related diseases:
First, 120 176 PubMed abstracts and 180 063 sentences related to immunosuppression were collected with these keywords: ‘immunosuppression,’ ‘tumor immune escape,’ ‘inhibit T cell,’ ‘immunotolerance,’ ‘immune checkpoint,’ ‘immune deviation,’ ‘peripheral tolerance,’ ‘central tolerance,’ ‘allograft tolerance,’ ‘oral tolerance,’ ‘split tolerance,’ ‘self-tolerance,’ ‘natural tolerance,’ ‘acquired tolerance’ and their lexical variants.
Second, 3634 candidate human immunosuppression genes were recognized and extracted from these sentences based on PR (14 (link)) which co-occurred with the immunosuppression keywords at single-sentence level. That is to say, a gene occurs together with at least one of the immunosuppression keywords in a single sentence.
Third, three-round strict manual curation was performed on these candidates by our experts generating 995 high confidence human immunosuppression genes:
Round 1: All candidate immunosuppression genes and supporting evidence were checked by two experienced researchers independently.
Round 2: These selected genes and supporting evidence were submitted to the internal reviewer team, in which all immunosuppression genes were manually reviewed by three experts.
Round 3: All co-authors were asked to randomly check immunosuppression genes from our website to make sure that all immunosuppression genes stored in our database are of high confidence. Each co-author randomly checked 200 immunosuppression genes and 99.5% of them are correct on average.
Disease terms were also extracted from these abstracts based on Human Disease Ontology (DO) (15 (link)). Associations between immunosuppression genes/proteins and human diseases were identified based on single-sentence level co-occurrence. Furthermore, among these selected genes, those with the function of immune checkpoint were recognized through manual curation. A full list of immunosuppression-related membrane proteins is established based on the Gene Ontology Annotation (UniProt-GOA) Database (16 (link)) for the discovery of promising immune checkpoints as most of the immune checkpoints are membrane proteins.
Immunosuppression gene related drugs were extracted based on Drugbank (17 (link)). First, we extracted 410 drugs under ‘Immunoglobulins,’ ‘Immunoproteins,’ ‘Immunosuppressive Agents,’ ‘Antineoplastic and Immunomodulating Agents’ categories from Drugbank and then we mapped these drugs to immunosuppression genes using disease-drug relations extracted from the XML format of Drugbank. Finally, Manual validation was performed on these mappings to ensure data quality and 270 immunosuppression gene related drugs were obtained.

Based on the fact that the interactive compounds often involve in similar biological activities [11] (link), it is feasible to predict the ATC-class of a query drug using the information of chemical-chemical interactions, as described below.
STITCH (Search tool for interactions of chemicals) [19] (link) is a large database containing known and predicted interactions between chemicals and between proteins derived from experiments, literature and other databases. We downloaded the information of chemical-chemical interactions from http://stitch.embl.de:8080/download/chemical_chemical.links.v2.0.tsv.gz. Each of these interactions was evaluated by a confidence score, ranging from 1 to 1000, to reflect the likelihood of its occurrence. For any two drugs d₁ and d₂, their interaction confidence score was denoted by . Particularly, if the interaction between d₁ and d₂ does not exist in STITCH, their interaction confidence score was set as zero, i.e., .
Suppose that a training dataset consists of n drugs , and that the 14 main ATC-classes are denoted by , where C₁ represents “Alimentary tract and metabolism”, C₂ “Blood and blood forming organs”, and so forth (see Table 1). The ATC-classes of any drug d_i can be formulated as where According to the chemical-chemical interaction approach, the likelihood for a query drug belonging to C_j, denoted as , can be calculated by where means that is an element of the training dataset . According Eq.6, the likelihood that belongs to C_j can be formulated as the maximum of the interaction confidence scores between and those drugs that belong to C_j in the training dataset . Obviously, the larger the score is, the more likely that belongs to . When , it means that the probability for the drug belonging to the class C_j is zero. Given a query drug compound , suppose the outcome derived from Eq.6 is which means that the highest probability for the drug belonging to the ATC-class is (“Antineoplastic and immunomodulating agents”), followed by (“Alimentary tract and metabolism”), and so forth (cf. Table 1). If there is a tie between two terms in Eq.7, then the probabilities for the drug belonging to the two corresponding classes are the same. But this kind of tie case rarely happened.
Note that the outcome of Eq.6 might turn out to be trivial, i.e., indicating that no chemical-chemical interaction exists for the query drug in the training dataset ; i.e., Under such a circumstance, no meaningful result would be obtained by the “interaction-based” method, and we should instead use the “similarity-based method as described in the next section.

This single-center, retrospective study was conducted at the Department of Gastroenterological Surgery, University of Health Sciences Kosuyolu High Specialization Education and Research Hospital, Istanbul, Turkey. The study was carried out in accordance with the Helsinki Declaration and local laws and regulations. This study was approved by the ethical committee of Kosuyolu High Specialization Education and Research Hospital with an IRB number: 2020/14/404.
Between December 2006 and December 2019, medical records of 324 patients who underwent gastric cancer surgery were retrospectively reviewed and data of 163 eligible patients were enrolled in the study (Figure 1). Patients aged over 18 who underwent a curative surgery for TNM stage II or III GC were considered eligible for this study. All patients underwent open total or subtotal gastrectomy with D2 lymphadenectomy. Patients who underwent emergency surgery, had immunodeficiency or lymphoproliferative disease and had taken immunomodulatory drugs were excluded. Also, patients whose adjuvant chemotherapy was not completed were not included into the study.
Data regarding the patients' age, gender, comorbidity status, presence/absence of lymphovascular and perineural invasions (LVI and PNI), tumor histological grade, tumor size and location, total number of harvested LNs and metastatic LNs, size of the largest MLN, length of hospital stay, postoperative complications, overall survival (OS), neoadjuvant treatment status were recorded. The Clavien-Dindo classification was used to analyze postoperative complications, and grade III or higher complications were defined as major complications (11 ).
Adjuvant chemotherapy was given to all patients with a pathological stage II and III gastric cancer with LN metastases. DCF (Docetaxel, cisplatin, 5-fluorouracil) or FLOT (5-fluorouracil, leucovorin, oxaliplatin, docetaxel) regimens were given as both neoadjuvant and adjuvant chemotherapy.
The software IBM® SPSS® (Statistical Package for the Social Sciences) version 23 (IBM Corp. Armonk, NY, USA) was used for statistical analysis. Qualitative data were presented as frequency and percentage. The distribution of numerical data was performed using the Kolmogorov–Smirnov test with the non-normal distribution results. Quantitative data were given as median with Interquartile Range (IQR). The association of major complications and survival with categorical variables was analyzed using Chi-square, Fisher's exact tests, and Likelihood ratio. The Mann–Whitney-U test was used to examine whether major complications and survival were related to age, metastatic lymph node size, and length of hospital stay. The Kaplan–Meier method and the log-rank test were used to conduct the survival analyses of the metastatic lymph node size. Further, multivariate Cox regression analyses were performed to examine role of the metastatic lymph node size in predicting mortality. A p-value of less than 0.05 was defined as statistically significant.