PharmaNet: Pharmaceutical discovery with deep recurrent neural networks

The discovery and development of novel pharmaceuticals is an area of active research mainly due to the large investments required and long payback times. As of 2016, the development of a novel drug candidate required up to $ USD 2.6 billion in investment for only 10% rate of approval by the FDA. To help decreasing the costs associated with the process, a number of in silico approaches have been developed with relatively low success due to limited predicting performance. Here, we introduced a machine learning-based algorithm as an alternative for a more accurate search of new pharmacological candidates, which takes advantage of Recurrent Neural Networks (RNN) for active molecule prediction within large databases. Our approach, termed PharmaNet was implemented here to search for ligands against specific cell receptors within 102 targets of the DUD-E database, which contains 22886 active molecules. PharmaNet comprises three main phases. First, a SMILES representation of the molecule is converted into a raw molecular image. Second, a convolutional encoder processes the data to obtain a fingerprint molecular image that is finally analyzed by a Recurrent Neural Network (RNN). This approach enables precise predictions of the molecules’ target on the basis of the feature extraction, the sequence analysis and the relevant information filtered out throughout the process. Molecule Target prediction is a highly unbalanced detection problem and therefore, we propose that an adequate evaluation metric of performance is the area under the Normalized Average Precision (NAP) curve. PharmaNet largely surpasses the previous state-of-the-art method with 97.7% in the Receiver Operating Characteristic curve (ROC-AUC) and 65.5% in the NAP curve. We obtained a perfect performance for human farnesyl pyrophosphate synthase (FPPS), which is a potential target for antimicrobial and anticancer treatments. We decided to test PharmaNet for activity prediction against FPPS by searching in the CHEMBL data set. We obtained three (3) potential inhibitors that were further validated through both molecular docking and in silico toxicity prediction. Most importantly, one of this candidates, CHEMBL2007613, was predicted as a potential antiviral due to its involvement on the PCDH17 pathway, which has been reported to be related to viral infections.

Analytical methods for assessing changes induced by gamma exposure in an animal model

SUMMARY OBJECTIVE: Ionizing radiation can cause radio-induced changes in the cellular metabolome due to the breakdown of DNA bonds. Our goal was to find the early tissue response to radiation exposure supported by distinct analytical methods. METHODS: Histological analyses were performed on the organs extracted from rats to search for microscopic changes. The histological slides stained with hematoxyline-eosin (HE) were analyzed in magnification (40x). Subsequently, the tissues were subjected to mass spectrometry that allowed molecular analysis and DESI-MSI that generated the molecular image of lipids, assessing changes in intensities, especially in the brain. RESULTS: The histological analysis found nonspecific inflammatory changes; no areas of fibrosis, necrosis, or apoptosis were identified, suggesting non-morphological tissue alterations. However, the DESI-MSI images of brain lipids allowed the observation of many radio-induced changes in the lipid's intensities. CONCLUSIONS: No early radio induced histological or mass weight changes in the radiation exposed rats could be observed at 5 Gy. However, early changes in the molecular level were observed in the DESI-MSI images of the brain lipids. The DESI-MSI method proved to be efficient and relevant, allowing a regional molecular analysis of the tissues, expanding a new field of study that is still in its infancy: radiometabolomics.

IMG-05. INITIAL RADIOGRAPHIC ASSESSMENT OF DWI AND ADC VALUES IN CHILDREN AND YOUNG ADULTS TREATED WITH DAY101 (TAK-580) FOR RECURRENT LOW-GRADE GLIOMAS (LGG) HARBORING MAPK ALTERATIONS

BACKGROUND Apparent diffusion coefficient (ADC) is a quantitative measure reflecting observed net movement of water calculated from a diffusion-weighted image (DWI), correlating with tumor cellularity. The higher cellularity of high-grade gliomas results in diffusion restriction and reduced ADC values, whereas the lower cellularity of low-grade gliomas (LGGs) gives higher ADC values. Here we examine changes in ADC values in patients with LGGs treated with the type 2 RAF inhibitor DAY101 (formerly TAK580). METHODS Historical, baseline, and on-treatment brain MRIs for 9 patients enrolled on a phase 1 study of DAY101 in children and young adults with radiographically recurrent or progressive LGG harboring MAPK pathway alterations were obtained, de-identified and independently evaluated for ADC changes. Time points included baseline, first follow-up, and best response. Data processing of ADC estimates was performed using pmod molecular image software package. ADC changes were displayed as a histogram with mean values. Results were based upon a single read paradigm. RESULTS There was a clear shift to lower ADC values for the solid component of tumors, reflecting changes in cellularity and tissue organization, while necrosis correlated with a shift toward higher ADC values. DWI reveals reduced ADCs in responding tumors, with the percent change in ADC from baseline correlating with deeper RANO responses. CONCLUSION DWI analysis reveals reductions in ADC values that correlates with treatment response and a shift toward more normal cellularity in tumors treated with DAY101. Changes in ADC may represent a novel imaging biomarker, reflecting biological response to DAY101 treatment.

BIOM-37. INITIAL RADIOGRAPHIC ASSESSMENT OF ADC VALUES IN PEDIATRIC PATIENTS TREATED WITH DAY101 FOR RECURRENT OR PROGRESSIVE LOW-GRADE GLIOMAS (LGGs) HARBORING MEK/ERK PATHWAY ALTERATIONS

BACKGROUND Calculated from a diffusion-weight image (DWI), the apparent diffusion coefficient (ADC) is a quantitative measure that reflects observed net movement of water and correlates to tumor cellularity. We examine the changes in ADC values in patients with LGG treated with dimeric, pan-RAF inhibitor DAY101 (formerly TAK-580/MLN2480). METHODS Focusing on ADC change with treatment, we reviewed historical, baseline, and on-treatment brain MRIs for 9 patients enrolled on our institutional, IRB-approved phase 1 trial of DAY101 in children and young adults with radiographically recurrent or progressive LGG harboring MEK/ERK pathway alterations. De-identified DICOM MRI files were independently reviewed. Time points selected included baseline, first follow-up and best response. The pmod molecular image software package was utilized for data processing of ADC estimates. ADC changes were displayed as a histogram with mean values. Results were based upon a single read paradigm. RESULTS A shift to lower ADC values for solid components of these tumors was observed, reflecting cellularity and organization; while necrosis correlated with a shift toward higher ADC values. DWI results show reduced ADCs in responding patients, with greater percent change in ADC from baseline associated with deeper responses among treated patients. DWI for treated patients with resultant stable disease showed no significant change in ADC values from baseline while a shift toward higher ADC values was evident in treated, progressing tumors. CONCLUSION Preliminary DWI analysis reveals that a reduction in ADC values may correlate with treatment response and a shift toward more normal cellularity in tumors treated with DAY101. This method will be applied to a larger cohort of patients in an ongoing phase 1 trial (NCT03429803) and a planned phase 2 trial.

PharmaNet: Pharmaceutical discovery with deep recurrent neural networks

The discovery and development of novel pharmaceuticals is an area of active research mainly due to the large investments required and long payback times. As of 2016, the development of a novel drug candidate required up to $ USD 2.6 billion in investment for only 10% rate of approval by the FDA. To help decreasing the costs associated with the process, a number of in silico approaches have been developed with relatively low success due to limited predicting performance. Here, we introduced a machine learning-based algorithm as an alternative for a more accurate search of new pharmacological candidates, which takes advantage of Recurrent Neural Networks (RNN) for active molecule prediction within large databases. Our approach, termed PharmaNet was implemented here to search for ligands against specific cell receptors within 102 targets of the DUD-E database, which contains 22886 active molecules. PharmaNet comprises three main phases. First, a SMILES representation of the molecule is converted into a raw molecular image. Second, a convolutional encoder processes the data to obtain a fingerprint molecular image that is finally analyzed by a Recurrent Neural Network (RNN). This approach enables precise predictions of the molecules’ target on the basis of the feature extraction, the sequence analysis and the relevant information filtered out throughout the process. Molecule Target prediction is a highly unbalanced detection problem and therefore, we propose that an adequate evaluation metric of performance is the area under the Normalized Average Precision (NAP) curve. PharmaNet largely surpasses the previous state-of-the-art method with 95.8% in the Receiver Operating Characteristic curve (ROC-AUC) and 58.9% in the NAP curve. We obtained a perfect performance for human farnesyl pyrophosphate synthase (FPPS), which is a potential target for antimicrobial and anticancer treatments. We decided to test PharmaNet for activity prediction against FPPS by searching in the CHEMBL data set. We obtained [3] potential inhibitors that were further validated through both molecular docking and in silico toxicity prediction. Most importantly, one of this candidates, CHEMBL2007613, was predicted as a potential antiviral due to its involvement on the PCDH17 pathway, which has been reported to be related to viral infections.

Molecular Image-Based Prediction Models of Nuclear Receptor Agonists and Antagonists Using the DeepSnap-Deep Learning Approach with the Tox21 10K Library

The interaction of nuclear receptors (NRs) with chemical compounds can cause dysregulation of endocrine signaling pathways, leading to adverse health outcomes due to the disruption of natural hormones. Thus, identifying possible ligands of NRs is a crucial task for understanding the adverse outcome pathway (AOP) for human toxicity as well as the development of novel drugs. However, the experimental assessment of novel ligands remains expensive and time-consuming. Therefore, an in silico approach with a wide range of applications instead of experimental examination is highly desirable. The recently developed novel molecular image-based deep learning (DL) method, DeepSnap-DL, can produce multiple snapshots from three-dimensional (3D) chemical structures and has achieved high performance in the prediction of chemicals for toxicological evaluation. In this study, we used DeepSnap-DL to construct prediction models of 35 agonist and antagonist allosteric modulators of NRs for chemicals derived from the Tox21 10K library. We demonstrate the high performance of DeepSnap-DL in constructing prediction models. These findings may aid in interpreting the key molecular events of toxicity and support the development of new fields of machine learning to identify environmental chemicals with the potential to interact with NR signaling pathways.

P0354MATRIX ASSISTED LASER DESORPTION/IONIZATION MASS SPECTROMETRY IN PROTEIN ALTERATIONS ASSOCIATED WITH THE PROGRESSION OF IGA NEPHROPATHY DISCOVERY

Background and Aims IgA nephropathy (IgAN) is one of the most diffuse glomerulonephrites worldwide, but there are still many issues regarding its prognosis and pathogenesis understanding. Although the diagnosis is established by renal biopsy examination, there are still remaining pitfalls in primary origin discrimination, and therefore for prognosis and outcome for the patient. Method In this pilot study we performed matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging (MSI) with a high spatial resolution (raster sampling of 20 μm) on formalin fixed paraffin embedded renal biopsies obtained from patients with IgAN (n=11) and other mesangioproliferative glomerulonephrites (MesPGN, n=6) attempting to enlighten proteomic alterations that may be associated with the progression of IgAN. The patients in both disease groups were separated according to their CKD stage (CKDI, CKDII, CKDIII and more). After MALDI-MSI analysis, all biopsies were stained with hematoxylin and eosin, scanned and overlaid with molecular image in order to verify the signals’ spatial distribution. Results Using MALDI-MSI we detected clear differences in the proteomic profiles of IgAN and other MesPGN tissues. Fourteen signals (AUC ≥ 0.8) were observed to have an altered intensity among the different CKD stages within the IgAN group. In particular, large increases in the intensity of these signals could be observed at CKD stages II and above. Putatively identified, these signals primarily corresponded to proteins involved in inflammatory and healing pathways and their increased intensity was localised within regions of tissue with large amounts of inflammatory cells or sclerosis (Figure 1), verified by immunohistochemical staining (Figure 2). Conclusion The capability of MALDI-MSI to provide highly spatially resolved proteomics analysis of complex renal tissue demonstrates it a useful additional diagnostic and prognostic tool and a promising approach in the search for prognostic or predictive markers in glomerular diseases.