PaccMann: a web service for interpretable anticancer compound sensitivity prediction.

The identification of new targeted and personalized therapies for cancer requires the fast and accurate assessment of the drug efficacy of potential compounds against a particular biomolecular sample. It has been suggested that the integration of complementary sources of information might strengthen the accuracy of a drug efficacy prediction model. Here, we present a web-based platform for the Prediction of AntiCancer Compound sensitivity with Multimodal Attention-based Neural Networks (PaccMann). PaccMann is trained on public transcriptomic cell line profiles, compound structure information and drug sensitivity screenings, and outperforms state-of-the-art methods on anticancer drug sensitivity prediction. On the open-access web service (https://ibm.biz/paccmann-aas), users can select a known drug compound or design their own compound structure in an interactive editor, perform in-silico drug testing and investigate compound efficacy on publicly available or user-provided transcriptomic profiles. PaccMann leverages methods for model interpretability and outputs confidence scores as well as attention heatmaps that highlight the genes and chemical sub-structures that were more important to make a prediction, hence facilitating the understanding of the model's decision making and the involved biochemical processes. We hope to serve the community with a toolbox for fast and efficient validation in drug repositioning or lead compound identification regimes.

Hydrodynamics in Cell Studies.

Hydrodynamic phenomena are ubiquitous in living organisms and can be used to manipulate cells or emulate physiological microenvironments experienced in vivo. Hydrodynamic effects influence multiple cellular properties and processes, including cell morphology, intracellular processes, cell-cell signaling cascades and reaction kinetics, and play an important role at the single-cell, multicellular, and organ level. Selected hydrodynamic effects can also be leveraged to control mechanical stresses, analyte transport, as well as local temperature within cellular microenvironments. With a better understanding of fluid mechanics at the micrometer-length scale and the advent of microfluidic technologies, a new generation of experimental tools that provide control over cellular microenvironments and emulate physiological conditions with exquisite accuracy is now emerging. Accordingly, we believe that it is timely to assess the concepts underlying hydrodynamic control of cellular microenvironments and their applications and provide some perspective on the future of such tools in in vitro cell-culture models. Generally, we describe the interplay between living cells, hydrodynamic stressors, and fluid flow-induced effects imposed on the cells. This interplay results in a broad range of chemical, biological, and physical phenomena in and around cells. More specifically, we describe and formulate the underlying physics of hydrodynamic phenomena affecting both adhered and suspended cells. Moreover, we provide an overview of representative studies that leverage hydrodynamic effects in the context of single-cell studies within microfluidic systems.

Toward Explainable Anticancer Compound Sensitivity Prediction via Multimodal Attention-Based Convolutional Encoders.

In line with recent advances in neural drug design and sensitivity prediction, we propose a novel architecture for interpretable prediction of anticancer compound sensitivity using a multimodal attention-based convolutional encoder. Our model is based on the three key pillars of drug sensitivity: compounds' structure in the form of a SMILES sequence, gene expression profiles of tumors, and prior knowledge on intracellular interactions from protein-protein interaction networks. We demonstrate that our multiscale convolutional attention-based encoder significantly outperforms a baseline model trained on Morgan fingerprints and a selection of encoders based on SMILES, as well as the previously reported state-of-the-art for multimodal drug sensitivity prediction (R2 = 0.86 and RMSE = 0.89). Moreover, the explainability of our approach is demonstrated by a thorough analysis of the attention weights. We show that the attended genes significantly enrich apoptotic processes and that the drug attention is strongly correlated with a standard chemical structure similarity index. Finally, we report a case study of two receptor tyrosine kinase (RTK) inhibitors acting on a leukemia cell line, showcasing the ability of the model to focus on informative genes and submolecular regions of the two compounds. The demonstrated generalizability and the interpretability of our model testify to its potential for in silico prediction of anticancer compound efficacy on unseen cancer cells, positioning it as a valid solution for the development of personalized therapies as well as for the evaluation of candidate compounds in de novo drug design.

Predictive microfluidic control of regulatory ligand trajectories in individual pluripotent cells.

Local (cell-level) signaling environments, regulated by autocrine and paracrine signaling, and modulated by cell organization, are hypothesized to be fundamental stem cell fate control mechanisms used during development. It has, however, been challenging to demonstrate the impact of cell-level organization on stem cell fate control and to relate stem cell fate outcomes to autocrine and paracrine signaling. We address this fundamental problem using a combined in silico and experimental approach in which we directly manipulate, using laminar fluid flow, the local impact of endogenously secreted gp130-activating ligands and their activation of signal transducer and activator of transcription3 (STAT3) signaling in mouse embryonic stem cells (mESC). Our model analysis predicted that flow-dependent changes in autocrine and paracrine ligand binding would impact heterogeneity in cell- and colony-level STAT3 signaling activation and cause a gradient of cell fate determination along the direction of flow. Interestingly, analysis also predicted that local cell density would be inversely proportional to the degree to which endogenous secretion contributed to cell fate determination. Experimental validation using functional activation of STAT3 by secreted factors under microfluidic perfusion culture demonstrated that STAT3 activation and consequently mESC fate were manipulable by flow rate, position in the flow field, and local cell organization. As a unique demonstration of how quantitative control of autocrine and paracrine signaling can be integrated with spatial organization to elicit higher order cell fate effects, this work provides a general template to investigate organizing principles due to secreted factors.

PaccMannRL: De novo generation of hit-like anticancer molecules from transcriptomic data via reinforcement learning.

With the advent of deep generative models in computational chemistry, in-silico drug design is undergoing an unprecedented transformation. Although deep learning approaches have shown potential in generating compounds with desired chemical properties, they disregard the cellular environment of target diseases. Bridging systems biology and drug design, we present a reinforcement learning method for de novo molecular design from gene expression profiles. We construct a hybrid Variational Autoencoder that tailors molecules to target-specific transcriptomic profiles, using an anticancer drug sensitivity prediction model (PaccMann) as reward function. Without incorporating information about anticancer drugs, the molecule generation is biased toward compounds with high predicted efficacy against cell lines or cancer types. The generation can be further refined by subsidiary constraints such as toxicity. Our cancer-type-specific candidate drugs are similar to cancer drugs in drug-likeness, synthesizability, and solubility and frequently exhibit the highest structural similarity to compounds with known efficacy against these cancer types.

Controlled self-assembly of quantum dot-block copolymer colloids in multiphase microfluidic reactors.

The controlled self-assembly of large compound quantum dot micelles (QDCMs), consisting of constituents of polymer-stabilized quantum dots (QDs) and amphiphilic polystyrene-b-poly(acrylic acid) stabilizing chains, in gas-liquid segmented microfluidic reactors is demonstrated. Self-assembly is initiated by fast mixing of water with the polymer constituents via chaotic advection, as liquid plugs containing reactants move through a sinusoidal mixing channel. The resulting QDCMs are then processed within a postformation channel, where circulating flow patterns develop within the liquid plugs, followed by off-chip quenching and analysis by transmission electron microscopy (TEM). Particle processing via circulating flow is found to involve a combination of particle growth via collision-induced coalescence and shear-induced particle breakup. The final mean QDCM sizes represent kinetic states arising from the competition between these two mechanisms, depending on tunable chemical and flow parameters. A systematic investigation of the experimental variables that influence particle size and polydispersity, including water concentration, flow rate, and the gas-to-liquid flow ratio, is conducted, demonstrating tunability of QDCM sizes in the range of approximately 40-140 nm. The importance of shear-induced particle breakup in the limit of high shear is illustrated by a common minimum particle size, 41 +/- 1 nm, which is achieved for all water contents by increasing the total flow rate to sufficiently high values.

Network-based Biased Tree Ensembles (NetBiTE) for Drug Sensitivity Prediction and Drug Sensitivity Biomarker Identification in Cancer.

We present the Network-based Biased Tree Ensembles (NetBiTE) method for drug sensitivity prediction and drug sensitivity biomarker identification in cancer using a combination of prior knowledge and gene expression data. Our devised method consists of a biased tree ensemble that is built according to a probabilistic bias weight distribution. The bias weight distribution is obtained from the assignment of high weights to the drug targets and propagating the assigned weights over a protein-protein interaction network such as STRING. The propagation of weights, defines neighborhoods of influence around the drug targets and as such simulates the spread of perturbations within the cell, following drug administration. Using a synthetic dataset, we showcase how application of biased tree ensembles (BiTE) results in significant accuracy gains at a much lower computational cost compared to the unbiased random forests (RF) algorithm. We then apply NetBiTE to the Genomics of Drug Sensitivity in Cancer (GDSC) dataset and demonstrate that NetBiTE outperforms RF in predicting IC50 drug sensitivity, only for drugs that target membrane receptor pathways (MRPs): RTK, EGFR and IGFR signaling pathways. We propose based on the NetBiTE results, that for drugs that inhibit MRPs, the expression of target genes prior to drug administration is a biomarker for IC50 drug sensitivity following drug administration. We further verify and reinforce this proposition through control studies on, PI3K/MTOR signaling pathway inhibitors, a drug category that does not target MRPs, and through assignment of dummy targets to MRP inhibiting drugs and investigating the variation in NetBiTE accuracy.

Deep-Reaching Hydrodynamic Flow Confinement: Micrometer-Scale Liquid Localization for Open Substrates With Topographical Variations.

We present a method for nonintrusive localization and reagent delivery on immersed biological samples with topographical variation on the order of hundreds of micrometers. Our technique, which we refer to as the deep-reaching hydrodynamic flow confinement (DR-HFC), is simple and passive: it relies on a deep-reaching hydrodynamic confinement delivered through a simple microfluidic probe design to perform localized microscale alterations on substrates as deep as 600 µm. Designed to scan centimeter-scale areas of biological substrates, our method passively prevents sample intrusion by maintaining a large gap between the probe and the substrate. The gap prevents collision of the probe and the substrate and reduces the shear stress experienced by the sample. We present two probe designs: linear and annular DR-HFC. Both designs comprise a reagent-injection aperture and aspiration apertures that serve to confine the reagent. We identify the design parameters affecting reagent localization and depth by DR-HFC and study their individual influence on the operation of DR-HFC numerically. Using DR-HFC, we demonstrate localized binding of antihuman immunoglobulin G (IgG) onto an activated substrate at various depths from 50 to 600 µm. DR-HFC provides a readily implementable approach for noninvasive processing of biological samples applicable to the next generation of diagnostic and bioanalytical devices.

Bubble pump: scalable strategy for in-plane liquid routing.

We present an on-chip liquid routing technique intended for application in well-based microfluidic systems that require long-term active pumping at low to medium flowrates. Our technique requires only one fluidic feature layer, one pneumatic control line and does not rely on flexible membranes and mechanical or moving parts. The presented bubble pump is therefore compatible with both elastomeric and rigid substrate materials and the associated scalable manufacturing processes. Directed liquid flow was achieved in a microchannel by an in-series configuration of two previously described "bubble gates", i.e., by gas-bubble enabled miniature gate valves. Only one time-dependent pressure signal is required and initiates at the upstream (active) bubble gate a reciprocating bubble motion. Applied at the downstream (passive) gate a time-constant gas pressure level is applied. In its rest state, the passive gate remains closed and only temporarily opens while the liquid pressure rises due to the active gate's reciprocating bubble motion. We have designed, fabricated and consistently operated our bubble pump with a variety of working liquids for >72 hours. Flow rates of 0-5.5 µl min(-1), were obtained and depended on the selected geometric dimensions, working fluids and actuation frequencies. The maximum operational pressure was 2.9 kPa-9.1 kPa and depended on the interfacial tension of the working fluids. Attainable flow rates compared favorably with those of available micropumps. We achieved flow rate enhancements of 30-100% by operating two bubble pumps in tandem and demonstrated scalability of the concept in a multi-well format with 12 individually and uniformly perfused microchannels (variation in flow rate <7%). We envision the demonstrated concept to allow for the consistent on-chip delivery of a wide range of different liquids that may even include highly reactive or moisture sensitive solutions. The presented bubble pump may provide active flow control for analytical and point-of-care diagnostic devices, as well as for microfluidic cells culture and organ-on-chip platforms.

Shaken, and stirred: oscillatory segmented flow for controlled size-evolution of colloidal nanomaterials.

We introduce oscillatory segmented flow as a compact microfluidic format that accommodates slow chemical reactions for the solution-phase processing of colloidal nanomaterials. The strategy allows the reaction progress to be monitored at a dynamic range of up to 80 decibels (i.e., residence times of up to one day, equivalent to 720-14,400 times the mixing time) from only one sensing location. A train of alternating gas bubbles and liquid reaction compartments (segmented flow) was initially formed, stopped and then subjected to a consistent back-and-forth motion. The oscillatory segmented flow was obtained by periodically manipulating the pressures at the device inlet and outlet via square wave signals generated by non-wetted solenoid valves. The readily implementable format significantly reduced the device footprint as compared with continuous segmented flow. We investigated mixing enhancement for varying liquid segment lengths, oscillation amplitudes and oscillation frequencies. The etching of gold nanorods served as a case study to illustrate the utility of the approach for dynamic characterization and precise control of colloidal nanomaterial size and shape for 5 h. Oscillatory segmented flows will be beneficial for a broad range of lab-on-a-chip applications that require long processing times.

Bubble gate for in-plane flow control.

We introduce a miniature gate valve as a readily implementable strategy for actively controlling the flow of liquids on-chip, within a footprint of less than one square millimetre. Bubble gates provide for simple, consistent and scalable control of liquid flow in microchannel networks, are compatible with different bulk microfabrication processes and substrate materials, and require neither electrodes nor moving parts. A bubble gate consists of two microchannel sections: a liquid-filled channel and a gas channel that intercepts the liquid channel to form a T-junction. The open or closed state of a bubble gate is determined by selecting between two distinct gas pressure levels: the lower level corresponds to the "open" state while the higher level corresponds to the "closed" state. During closure, a gas bubble penetrates from the gas channel into the liquid, flanked by a column of equidistantly spaced micropillars on each side, until the flow of liquid is completely obstructed. We fabricated bubble gates using single-layer soft lithographic and bulk silicon micromachining procedures and evaluated their performance with a combination of theory and experimentation. We assessed the dynamic behaviour during more than 300 open-and-close cycles and report the operating pressure envelope for different bubble gate configurations and for the working fluids: de-ionized water, ethanol and a biological buffer. We obtained excellent agreement between the experimentally determined bubble gate operational envelope and a theoretical prediction based on static wetting behaviour. We report case studies that serve to illustrate the utility of bubble gates for liquid sampling in single and multi-layer microfluidic devices. Scalability of our strategy was demonstrated by simultaneously addressing 128 bubble gates.

Formation and shear-induced processing of quantum dot colloidal assemblies in a multiphase microfluidic chip.

The controlled self-assembly of polymer-stabilized quantum dots (QDs) into mesoscale aqueous spherical assemblies termed quantum dot compound micelles (QDCMs) using a two-phase gas-segmented microfluidic reactor is described. Self-assembly is initiated by the fast mixing of water (approximately 1 s) with a blend solution of polystyrene-coated QDs and amphiphilic polystyrene-block-poly(acrylic acid) stabilizing chains via chaotic advection within liquid plugs moving through a sinusoidal channel. Subsequent recirculating flow within a post-formation channel subjects the dynamic QDCMs to shear-induced processing, controlled via the flow rate and channel length, before a final quench into pure water. During processing, larger QDCMs within the initial population undergo breakup into smaller particles, resulting in smaller mean particle sizes, smaller relative standard deviations, and more skewed distribution shapes, as the overall shear exposure is increased. For these cases, shear-induced size reduction is sufficient to dominate surface tension-driven growth.