Evaluating Deep Learning models for predicting ALK-5 inhibition

Computational methods have been widely used in drug design. The recent developments in machine learning techniques and the ever-growing chemical and biological databases are fertile ground for discoveries in this area. In this study, we evaluated the performance of Deep Learning models in comparison to Random Forest, and Support Vector Regression for predicting the biological activity (pIC50) of ALK-5 inhibitors as candidates to treat cancer. The generalization power of the models was assessed by internal and external validation procedures. A deep neural network model obtained the best performance in this comparative study, achieving a coefficient of determination of 0.658 on the external validation set with mean square error and mean absolute error of 0.373 and 0.450, respectively. Additionally, the relevance of the chemical descriptors for the prediction of biological activity was estimated using Permutation Importance. We can conclude that the forecast model obtained by the deep neural network is suitable for the problem and can be employed to predict the biological activity of new ALK-5 inhibitors.

HGG-01. ENTRECTINIB IN RECURRENT OR REFRACTORY SOLID TUMORS INCLUDING PRIMARY CNS TUMORS: UPDATED DATA IN CHILDREN AND ADOLESCENTS

STARTRK-NG (phase 1/2) is evaluating entrectinib, a CNS-penetrant oral, TRK/ROS1/ALK tyrosine kinase inhibitor, in patients <21 years with recurrent/refractory solid tumors, including primary CNS tumors. After determining the recommended dose, 550mg/m2/day, in all-comers, expansion cohorts with gene-fusion-positive CNS/solid tumors (NTRK1/2/3, ROS1) are being enrolled. As of 5Nov2019 (data cut-off), 39 patients (4.9m–20y; median 7y) have been evaluated for response, classified as complete (CR) or partial response (PR), stable (SD) or progressive disease (PD) using RANO (CNS), RECIST (solid tumors), or Curie score (neuroblastoma). Responses in patients with fusion-positive tumors were Investigator-assessed (BICR assessments are ongoing) and occurred at doses ≥400mg/m2. Best responses in fusion-positive CNS tumors (n=14) were: 4 CR (GKAP1-NTRK2, ETV6-NTRK3 [n=2], EML1-NTRK2); 5 PR (KANK1-NTRK2, GOPC-ROS1, ETV6-NTRK3, TPR-NTRK1, EEF1G-ROS1); 3 SD (BCR-NTRK2, ARHGEF2-NTRK1, KIF21B-NTRK1); 2 PD (PARP6-NTRK3, EML4-ALK); and in fusion-positive solid tumors (n=8) were: 3 CR (ETV6-NTRK3 [n=2], DCTN1-ALK); 5 PR (EML4-NTRK3, TFG-ROS1 [n=3], KIF5B-ALK). Responses (Investigator-assessed) in non-fusion tumors (n=17) were: 1 CR (ALK F1174L mutation), 3 SD, 10 PD, 3 no data/unevaluable. The objective response rate (CR+PR/total) in patients with fusion-positive tumors was 77% (17/22) versus 6% (1/17) in those with non-fusion tumors. All 39 patients experienced ≥1 adverse event (AE); the most frequent AEs included weight gain and anemia (both 48.7%); increased ALT, increased AST, cough and pyrexia (all 46.2%); increased creatinine and vomiting (both 43.6%); and bone fractures (n=10, in 9 patients). Entrectinib has produced striking, rapid, and durable responses in solid tumors with target gene fusions, especially high-grade CNS neoplasms.

Structure-Based Virtual Screening, Molecular Dynamics and Binding Free Energy Calculations of Hit Candidates as ALK-5 Inhibitors

Activin-like kinase 5 (ALK-5) is involved in the physiopathology of several conditions, such as pancreatic carcinoma, cervical cancer and liver hepatoma. Cellular events that are landmarks of tumorigenesis, such as loss of cell polarity and acquisition of motile properties and mesenchymal phenotype, are associated to deregulated ALK-5 signaling. ALK-5 inhibitors, such as SB505154, GW6604, SD208, and LY2157299, have recently been reported to inhibit ALK-5 autophosphorylation and induce the transcription of matrix genes. Due to their ability to impair cell migration, invasion and metastasis, ALK-5 inhibitors have been explored as worthwhile hits as anticancer agents. This work reports the development of a structure-based virtual screening (SBVS) protocol aimed to prospect promising hits for further studies as novel ALK-5 inhibitors. From a lead-like subset of purchasable compounds, five molecules were identified as putative ALK-5 inhibitors. In addition, molecular dynamics and binding free energy calculations combined with pharmacokinetics and toxicity profiling demonstrated the suitability of these compounds to be further investigated as novel ALK-5 inhibitors.

Association between gene fusions and anti-EGFR resistance signature in colorectal cancer.

3564 Background: Acquired resistance to anti-EGFR therapy in metastatic colorectal cancer (mCRC) has been characterized by a circulating tumor DNA (ctDNA) signature including any sub-clonal RAS mutation, co-existing RAS mutations, or co-existing EGFR mutations. Here, we investigated if fusions in ctDNA are associated with this anti-EGFR signature for CRC patients (pts). Methods: 4289 advanced stage CRC pts underwent molecular profiling using a plasma-based NGS assay that included FGFR2, FGFR3, RET, ALK, NTRK1, and ROS1 fusions. Available clinical histories were reviewed. Correlations between fusions and clinicopathological features were measured with a Fischer exact test. Relative frequencies of genomic alterations were compared between fusion-present vs -absent cases with an unpaired t-test. Clonality for a given alteration was called for a relative variant allele frequency (rVAF) > 50 %, while subclonal was defined as < 50% rVAF. Results: 44 unique fusions were detected in 40 (1.1%) of the 3808 pts with alterations present: RET (N = 16), FGFR3 (N = 12), ALK (N = 10), NTRK1 (N = 3), ROS1 (N = 2), and FGFR2 (N = 1). Relative to non-fusion variants detected, fusions were more likely to be subclonal (OR 8.2, p < 0.0001). Mutations associated with a previously reported anti-EGFR resistance were found in FGFR3 (7/12 pts), RET (7/15 pts) and ALK (5/10 pts). In fusion-present cases, co-existing RAS mutations were more likely to be subclonal than non-fusion cases (OR 14, p < 0.0001). EGFR mutations were more common in fusion present cases (OR 3.7, p = 0.0001) and predominantly subclonal (97%). EGFR mutations aggregated to ectodomain sites (85%) previously linked to acquired anti-EGFR resistance. For 27 pts with available clinical histories, 21 (78%) received anti-EGFR treatment prior to ctDNA testing. Conclusions: Actionable fusions using a ctDNA NGS assay were predominantly subclonal and co-existed with subclonal RAS and EGFR mutations. These clinicopathologic features are consistent with a previously validated signature linked to resistance to anti-EGFR therapies in CRC. We hypothesize that fusions may arise as a previously undescribed mechanism of acquired resistance to anti-EGFR therapies in CRC pts.