Explaining Black Box Drug Target Prediction Through Model Agnostic Counterfactual Samples.

Many high-performance DTA deep learning models have been proposed, but they are mostly black-box and thus lack human interpretability. Explainable AI (XAI) can make DTA models more trustworthy, and allows to distill biological knowledge from the models. Counterfactual explanation is one popular approach to explaining the behaviour of a deep neural network, which works by systematically answering the question "How would the model output change if the inputs were changed in this way?". We propose a multi-agent reinforcement learning framework, Multi-Agent Counterfactual Drug-target binding Affinity (MACDA), to generate counterfactual explanations for the drug-protein complex. Our proposed framework provides human-interpretable counterfactual instances while optimizing both the input drug and target for counterfactual generation at the same time. We benchmark the proposed MACDA framework using the Davis and PDBBind dataset and find that our framework produces more parsimonious explanations with no loss in explanation validity, as measured by encoding similarity. We then present a case study involving ABL1 and Nilotinib to demonstrate how MACDA can explain the behaviour of a DTA model in the underlying substructure interaction between inputs in its prediction, revealing mechanisms that align with prior domain knowledge.

Mitigating cold-start problems in drug-target affinity prediction with interaction knowledge transferring.

Predicting the drug-target interaction is crucial for drug discovery as well as drug repurposing. Machine learning is commonly used in drug-target affinity (DTA) problem. However, the machine learning model faces the cold-start problem where the model performance drops when predicting the interaction of a novel drug or target. Previous works try to solve the cold start problem by learning the drug or target representation using unsupervised learning. While the drug or target representation can be learned in an unsupervised manner, it still lacks the interaction information, which is critical in drug-target interaction. To incorporate the interaction information into the drug and protein interaction, we proposed using transfer learning from chemical-chemical interaction (CCI) and protein-protein interaction (PPI) task to drug-target interaction task. The representation learned by CCI and PPI tasks can be transferred smoothly to the DTA task due to the similar nature of the tasks. The result on the DTA datasets shows that our proposed method has advantages compared to other pre-training methods in the DTA task.

GEFA: Early Fusion Approach in Drug-Target Affinity Prediction.

Predicting the interaction between a compound and a target is crucial for rapid drug repurposing. Deep learning has been successfully applied in drug-target affinity (DTA)problem. However, previous deep learning-based methods ignore modeling the direct interactions between drug and protein residues. This would lead to inaccurate learning of target representation which may change due to the drug binding effects. In addition, previous DTA methods learn protein representation solely based on a small number of protein sequences in DTA datasets while neglecting the use of proteins outside of the DTA datasets. We propose GEFA (Graph Early Fusion Affinity), a novel graph-in-graph neural network with attention mechanism to address the changes in target representation because of the binding effects. Specifically, a drug is modeled as a graph of atoms, which then serves as a node in a larger graph of residues-drug complex. The resulting model is an expressive deep nested graph neural network. We also use pre-trained protein representation powered by the recent effort of learning contextualized protein representation. The experiments are conducted under different settings to evaluate scenarios such as novel drugs or targets. The results demonstrate the effectiveness of the pre-trained protein embedding and the advantages our GEFA in modeling the nested graph for drug-target interaction.

Using the IPTG-Inducible Pgrac212 Promoter for Overexpression of Human Rhinovirus 3C Protease Fusions in the Cytoplasm of Bacillus subtilis Cells.

Expression and secretion of recombinant proteins in the endotoxin-free bacterium, Bacillus subtilis, has been thoroughly studied, but overexpression in the cytoplasm has been limited to only a few proteins. Here, we used the robust IPTG-inducible promoter, Pgrac212, to overexpress human rhinovirus 3C protease (HRV3C) in the cytoplasm of B. subtilis cells. A novel solubility tag, the N-terminal domain of the lysS gene of B. subtilis coding for a lysyl-tRNA synthetase was placed at the N terminus with a cleavage site for the endoprotease HRV3C, followed by His-HRV3C or His-GST-HRV3C. The recombinant protease was purified by using a Ni-NTA column. In this study, the His-HRV3C and His-GST-HRV3C proteases were overexpressed in the cytoplasm of B. subtilis at 11% and 16% of the total cellular proteins, respectively. The specific protease activities were 8065 U/mg for His-HRV3C and 3623 U/mg for His-GST-HRV3C. The purified enzymes were used to cleave two different substrates followed by purification of the two different protein targets, the green fluorescent protein and the beta-galactosidase. In conclusion, the combination of an inducible promoter Pgrac212 and a solubility tag allowed the overexpression of the HRV3C protease in the cytoplasm of B. subtilis. The resulting fusion protein was purified using a nickel column and was active in cleaving target proteins to remove the fusion tags. This study offers an effective method for producing recombinant proteins in the cytoplasm of endotoxin-free bacteria.

Development of inducer-free expression plasmids based on IPTG-inducible promoters for Bacillus subtilis.

BACKGROUND: Besides Escherichia coli, Bacillus subtilis is an important bacterial species for the production of recombinant proteins. Recombinant genes are inserted into shuttle expression vectors which replicate in both E. coli and in B. subtilis. The ligation products are first transformed into E. coli cells, analyzed for correct insertions, and the correct recombinant plasmids are then transformed into B. subtilis. A major problem using E. coli cells can be the strong basal level of expression of the recombinant protein which may interfere with the stability of the cells. To minimize this problem, we developed strong expression vectors being repressed in E. coli and inducer-free in B. subtilis. RESULTS: In general, induction of IPTG-inducible expression vectors is determined by the regulatory lacI gene encoding the LacI repressor in combination with the lacO operator on the promoter. To investigate the inducer-free properties of the vectors, we constructed inducer-free expression plasmids by removing the lacI gene and characterized their properties. First, we examined the ability to repress a reporter gene in E. coli, which is a prominent property facilitating the construction of the expression vectors carrying a target gene. The ß-galactosidase (bgaB gene) basal levels expressed from Pgrac01-bgaB could be repressed at least twice in the E. coli cloning strain. Second, the inducer-free production of BgaB from four different plasmids with the Pgrac01 promoter in B. subtilis was investigated. As expected, BgaB expression levels of inducer-free constructs are at least 37 times higher than that of the inducible constructs in the absence of IPTG, and comparable to those in the presence of the inducer. Third, using efficient IPTG-inducible expression vectors containing the strong promoter Pgrac100, we could convert them into inducer-free expression plasmids. The BgaB production levels from the inducer-free plasmid in the absence of the inducer were at least 4.5 times higher than that of the inducible vector using the same promoter. Finally, we used gfp as a reporter gene in combination with the two promoters Pgrac01 and Pgrac100 to test the new vector types. The GFP expression levels could be repressed at least 1.5 times for the Pgrac01-gfp+ inducer-free construct in E. coli. The inducer-free constructs Pgrac01-gfp+ and Pgrac100-gfp+ allowed GFP expression at high levels from 23 × 104 to 32 × 104 RFU units and 9-13% of total intracellular proteins. We could reconfirm the two major advantages of the new inducer-free expression plasmids: (1) Strong repression of the target gene expression in the E. coli cloning strain, and (2) production of the target protein at high levels in B. subtilis in the absence of the inducer. CONCLUSIONS: We propose a general strategy to generate inducer-free expression vector by using IPTG-inducible vectors, and more specifically we developed inducer-free expression plasmids using IPTG-inducible promoters in the absence of the LacI repressor. These plasmids could be an excellent choice for high-level production of recombinant proteins in B. subtilis without the addition of inducer and at the same time maintaining a low basal level of the recombinant proteins in E. coli. The repression of the recombinant gene expression would facilitate cloning of genes that potentially inhibit the growth of E. coli cloning strains. The inducer-free expression plasmids will be extended versions of the current available IPTG-inducible expression vectors for B. subtilis, in which all these vectors use the same cognate promoters. These inducer-free and previously developed IPTG-inducible expression plasmids will be a useful cassette to study gene expression at a small scale up to a larger scale up for the production of recombinant proteins.

Paris saponin II of Rhizoma Paridis--a novel inducer of apoptosis in human ovarian cancer cells.

Rhizoma Paridis (dried root and rhizome) has been an essential ingredient in traditional Chinese herbal medicine. In the past decade, active components of Rhizoma Paridis - the Paris saponins have emerged as promising anti-cancer agents. Among these saponins, polyphyllin D (Paris saponin (PS) I), has been extensively studied and proposed to be a potent antitumor agent. In this study, we continue to establish the efficacy and mechanisms underlying the cytotoxic effects of the steroidal PS members, namely formosanin C (PSII) in ovarian cancer treatment. We isolated PSII and evaluated its effects on a panel of ten human cell lines. Isolated PSII has potent inhibitory effects on the growth of tumor cells without deleterious effects to different normal cell types or benign neoplastic derived cells. While PSII, PSI, and etoposide are effective promoting agents for cell cycle arrest and apoptosis, PSII appeared to be marginally more potent than the later two in inhibiting SKOV3 cell growth. In PSII-treated SKOV3 cells, there was an elevation in proapoptotic elements including Bax, cytosolic cytochrome c, activated-caspase-3, and activated-caspase-9. The treatment also reduced extracellular signal-regulated kinase (ERK1/2) phosphorylation and anti-apoptotic Bcl-2 expression. We also assessed the antitumor efficacy of intraperitoneal administration of PSII in human SKOV3 ovarian cancer xenografts in athymic mice. PSII treatment significantly inhibited the growth of xenograft tumors relative to controls by 70% (p < 0.05). These findings demonstrated that, in addition to the unique selectivity against cancer cells, PSII is a potent antitumor molecule that may be developed as a cancer therapeutic agent.

Endostatin induces autophagy in endothelial cells by modulating Beclin 1 and beta-catenin levels.

Endostatin is a well-characterized endogenous inhibitor of angiogenesis that affects cell proliferation and migration by inhibiting integrin and Wnt-mediated signalling pathways. Here, we show that endothelial cells treated with native and P125A-endostatin activate autophagy. Because autophagy can either be protective or induce programmed cell death, experiments were carried out to understand the signalling pathways leading to autophagy in endothelial cells. P125A-endostatin treatment increased the levels of Beclin 1, a crucial molecule in vesicle nucleation and autophagy. The treatment also reduced the levels of Bcl-2, Bcl-x(L) and beta-catenin; however, progressively increasing amounts of Bcl-2 and Bcl-x(L) were found to be complexed with Beclin 1. Increased beta-catenin and Wnt-mediated signalling reduced Beclin 1 levels and rescued endothelial cells from endostatin-induced autophagy. Finally, knocking down Beclin 1 levels by RNA interference decreased autophagy and accelerated caspase activation in endostatin-treated cells. These studies suggest that endothelial cells may initiate autophagy as a survival response to limit the effects of angiogenesis inhibitors. Thus, interfering with autophagy can potentiate the effects of endostatin by promoting a switch to apoptosis.

Autophagy and angiogenesis inhibition.

Angiogenesis, the process by which new blood vessels are formed is critical for embryonic development and physiological functioning of normal tissues. Angiogenesis also plays a critical role in the pathology of many diseases including cancer, wherein the supply and demand for blood vessels determines the rate of cancer growth. A number of therapeutic strategies are being developed to inhibit pathological angiogenesis. Kringle domains of plasminogen such as kringle 5 (K5) and a proteolytic fragment of collagen type XVIII (endostatin) are well-characterized, potent angiogenesis inhibitors. These inhibitors activate different intracellular signaling pathways to induce apoptosis and inhibit cell proliferation. Recent studies from our group have shown that K5 and endostatin can also induce autophagy in addition to apoptosis in endothelial cells. A common feature of the two treatments was the upregulation of Beclin 1 levels leading to alterations in the Beclin 1-Bcl-2 complex. Angiogenesis inhibitor-induced autophagy in endothelial cells was independent of nutritional or hypoxic stress and initiated even in the presence of endothelial-specific survival factors such as vascular endothelial growth factor (VEGF). Interfering with the autophagic response by knocking down Beclin 1 levels dramatically increased apoptosis of endothelial cells. These findings identify the autophagic response as a novel target for enhancing the therapeutic efficacy of angiogenesis inhibitors.

Kringle 5 of human plasminogen, an angiogenesis inhibitor, induces both autophagy and apoptotic death in endothelial cells.

Inhibition of endothelial cell proliferation and angiogenesis is emerging as an important strategy in cancer therapeutics. Kringle 5 (K5) of human plasminogen is a potent angiogenesis inhibitor. Previous studies have shown K5 exposure promotes caspase activity and apoptosis in endothelial cells. Here we report that K5 treatment evokes an autophagic response in endothelial cells that is specific and initiated even in the absence of nutritional stress. Endothelial cells exposed to K5 up-regulated Beclin 1 levels within a few hours. Furthermore, progressively increasing amounts of antiapoptotic Bcl-2 were found to be complexed with Beclin 1, although total levels of Bcl-2 remained unchanged. Prolonged exposure to K5 ultimately led to apoptosis via mitochondrial membrane depolarization and caspase activation in endothelial cells. Knocking down Beclin 1 levels by RNA interference decreased K5 induced autophagy but accelerated K5-induced apoptosis. These studies suggest that interfering with the autophagic survival response can potentiate the antiangiogenic effects of Kringle 5 in endothelial cells.

Adeno-associated virus-mediated delivery of a mutant endostatin in combination with carboplatin treatment inhibits orthotopic growth of ovarian cancer and improves long-term survival.

A human ovarian cancer cell line, which migrates to mouse ovaries and establishes peritoneal carcinomatosis, was used to evaluate the cooperative effect of an antiangiogenic gene therapy combined with chemotherapy. The ovarian carcinoma cell line MA148 was genetically modified by "Sleeping Beauty" transposon-mediated delivery of DsRed2 fluorescent protein. Stable, high-level expression of DsRed protein enabled in vivo imaging of peritoneal dissemination of ovarian cancer. Both external and internal imaging, along with histopathology, showed migration of i.p. injected human ovarian cancer cell line to mouse ovaries. Using this model, we evaluated the effect of adeno-associated virus (AAV)-mediated expression of a mutant endostatin either alone or in combination with carboplatin treatment. A single i.m. injection of recombinant AAV (rAAV)-mutant human endostatin with P125A substitution (P125A-endostatin) showed sustained expression of mutant endostatin. Antiangiogenic gene therapy inhibited orthotopic growth of ovarian cancer and resulted in 33% long-term tumor-free survival. A single cycle of carboplatin treatment combined with mutant endostatin gene therapy resulted in 60% of the animals remaining tumor free for >200 days, which was significantly better than rAAV-LacZ and/or carboplatin. Combination treatment delayed tumor appearance in 40% of the animals, wherein the residual tumors were smaller in size with limited or no peritoneal metastasis. These studies suggest that AAV-mediated gene therapy of P125A-endostatin in combination with carboplatin is a useful method to inhibit peritoneal dissemination of ovarian carcinoma.