Automated solubility screening platform using computer vision

•We demonstrate a modular, closed-loop robotic platform for solubility screening•Automated solvent titration is informed by computer vision and turbidity monitoring•No human intervention or HPLC analysis required during the experimental loop•Solubility values obtained system match those via traditional methods Solubility screening an essential, routine process that often labor intensive. Robotic platforms have been developed to automate some aspects of manual involved. However, many existing systems rely on analytic techniques such as high-performance liquid chromatography, which require pre-calibration each compound can be resource consuming. In addition, automation not typically end-to-end, requiring user move vials, establish analytical interpret raw data. We closed-loop, flexible with integrated solid dosing capabilities relies iterative feedback successfully measure caffeine in multiple solvents. After initial researcher input (<2 min), ran autonomously, five different (20-80 min each). The resulting matched using techniques. defined amount will dissolve certain at given temperature when equilibrium. ability reliably screen fundamental areas chemical research industry sectors, from pharmaceutical applications (Coltescu et al., 2020Coltescu A.-R. Butnariu M. Sarac I. importance new drug Molecules.Biomed. Pharmacol. J. 2020; 13: 577-583Crossref Google Scholar; Hansen, 1969Hansen C.M. universality parameter.Product R&D. 1969; 8: 2-11Crossref Scopus (526) Scholar), where used development selection, material science, it limits formulation processing parameters (Hansen, 2004Hansen 50 Years parameters—past future.Prog. Org. Coat. 2004; 51: 77-84Crossref (223) Machui 2012Machui F. Langner S. Zhu X. Abbott Brabec C.J. Determination P3HT:PCBM binary gradient method: impact photovoltaic performance.Solar Energ. Mater. solar Cell. 2012; 100: 138-146Crossref (193) Walker 2011Walker B. Tamayo A. Duong D.T. Dang X.-D. Kim C. Granstrom Nguyen T.-Q. A systematic approach selection based cohesive energy densities Molecular bulk heterojunction system.Adv. Energy 2011; 1: 221-229Crossref (148) Scholar). Common determine solubility, broadly categorized ‘excess solid’ solvent’ methods, researchers perform incremental, tasks must adjusted repeated solute-solvent (Black 2013Black L. Liu Wei H. On measurement solubility.Org. Process. Res. Dev. 2013; 17: 486-492Crossref (24) case excess solid, solute added solution until undissolved solids are observed sample. allowed equilibrate long incubation time then sampled analyzed composition variety (e.g., gravimetric titration, UV-spectrometry, most commonly chromatography HPLC). method filtration, also called ‘shake-plate’ ‘slurry’, known provide highest quality requires additional step filtration at-temperature centrifugation prior (Selekman 2017Selekman J.A. Qiu Tran K. Stevens Rosso V. Simmons E. Xiao Y. Janey High-throughput development.Annu. Rev. Chem. Biomol. Eng. 2017; 525-547Crossref PubMed (36) adds complexity slows down workflow (Hoelke 2009Hoelke Gieringer Arlt Saal Comparison nephelometric, UV-spectroscopic, high-throughput determination aqueous Microtiter plates.Anal. (Washington). 2009; 81: 3165-3172Crossref (49) method, titration-like procedure applied: stepwise fully dissolved state observed. dissolution point measured light scattering imaging-based technology (Reus 2015Reus M.A. van der Heijden A.E.D.M. ter Horst J.H. clear points upon addition.Org. 2015; 19: 1004-1011Crossref (22) therefore obviate need pre-calibrated expensive like HPLC, but comes cost adding somewhat time-consuming addition actions. Recent advances enabled automated (Burger 2020Burger Maffettone P.M. Gusev V.V. Aitchison Bai Wang Li Alston B.M. Clowes R. al.A mobile chemist.Nature (London). 583: 237-241Crossref (92) 2018Li Lu Xu Tu Ye Xie Qian AIR-chem: Authentic intelligent robotics chemistry.J. Phys. 2018; 122: 9142-9148Crossref (21) MacLeod 2020MacLeod B.P. Parlane F.G.L. Morrissey T.D. Häse Roch L.M. Dettelbach K.E. Moreira Yunker L.P.E. Rooney M.B. Deeth J.R. al.Self-driving laboratory accelerated discovery thin-film materials.Sci. Adv. 6: eaaz8867Crossref (52) Scholar) aid decreasing hands-on increasing throughput While effective applied (Goodwin, 2006Goodwin J.J. Rationale benefit high screens discovery.Drug Discov. Today. Tech. 2006; 3: 67-71Crossref (13) Selekman Sou Bergström, 2018Sou T. Bergström C.A.S. Automated assays thermodynamic (equilibrium) determination.Drug 27: 11-19Crossref only partially still interpretation stages workflow. these continue well calibration screen. vast majority common instrument strategies (Alsenz 2007Alsenz Meister E.v. a. Haenel Development (PASS) assay early development.J. Pharm. Sci. 2007; 96: 1748-1762Abstract Full Text PDF (50) Chen 2002Chen Shen Evaluation multi-wavelength UV plate reader.Comb. High Throughput Screen. 2002; 5: 575-581Crossref (68) 2020Qiu Patel J.M. salt synthetic intermediates: effects solvents, counterions, counterion 24: 1262-1270Crossref (4) Tan 2005Tan Semin D. Wacker Cheetham An equilibrium enabling SPR study lead optimization.JALA. 2005; 10: 364-373Google Wenlock 2011Wenlock M.C. Austin R.P. Potter Barton P. highly determining compounds.J. Lab. Automation. 16: 276-284Crossref (20) reliable standard quantifying solution, has several drawbacks – significant context calibration. Moreover, chromophore present set test materials spent developing customized necessary accurate analysis. Other situ For example, Dinter al. (Dinter 2005Dinter Schuetz Blume Weinmann Harre Neh probe.J. 408-411Crossref (10) reported fiberoptic probe qualitative measurements, categorizing soluble, turbid uncertain within range. groups reader measurements both absorption modes (Chen Hoelke Dehring (Dehring 2004Dehring K.A. Workman H.L. Miller K.D. Mandagere Poole S.K. handling/laser-based nephelometry kinetic solubility.J. Biomed. Anal. 36: 447-456Crossref (71) uses laser-based solubility. visual checks. Machine offers minimally invasive direct means assessment extensively wide range chemistry thin film defect detection (Taherimakhsousi 2020Taherimakhsousi N. Booker E.P. Berlinguette C.P. Quantifying defects films machine vision.NPJ Comput. 111Crossref (1) lab object recognition (Eppel 2020Eppel Bismuth Aspuru-Guzik Computer vessels settings vector-LabPics data set.ACS Cent. 1743-1752Crossref (2) studies. Stukelj (Štukelj 2019aŠtukelj Svanbäck Agopov Löbmann Strachan Rades Yliruusi Direct amorphous solubility.Anal. 2019; 91: 7411-7417Crossref (7) Scholar, Štukelj 2019bŠtukelj Kristl Image-based investigation: biorelevant ? ? indomethacin.Anal. 3997-4003Crossref (8) 2020Štukelj Machine-vision-enabled analysis.Anal. 92: 9730-9738Crossref image-based single-particle-analysis Reus images taken ReactIR probe. less than controlled emission light, may due limited Here we automating combining (excess solvent) simple vision. Unlike make use this HPLC-free, relying solely recorded webcam custom algorithm monitor dissolved. platform's design purposefully its easy adaptation other workflows crystallization multi-solvent systems. accessed remote desktop security camera, easily allows live view experiment further remotely (see Table S1 Figure S7 Supplemental information (SI) document, details). Our consists basic modules: arm, system, (3) unit, stir plate, (5) (Figure 1A). Each module performs essential component testing: manipulating objects, controlling reaction conditions, recording study, respectively. All modules were connected computer. run Python master script, designed function independently, thereby reconfiguration expansion suit conditions. located atop deck 1B) equipped hardware components general-purpose tools help interact detail). N9 Selective Compliance Articulated Robot Arm (SCARA) rotatable gripper (North Robotics, www.northrobotics.com). It 3 elbows three dimensions. Both attached arm head handle specific equipment needed transfer, uncap, recap while needles dosing. Additional placed engaged experiments. Tecan Cavro syringe pump controller. One port reservoir probe, sampling any vial deck. contain solvent, acetonitrile our case, was prime tubing line hold 1 mL accurately deliver between 0.01 0.95 mL. Solvents vials capped pre-slit septa minimize evaporation contamination allowing needle end Refer Figures 1B. workflow, important able dose specified solid. on-deck extends platform. platform, Mettler Toledo Quantos powder modifications integrate dispensing into filling heads desired compound, all completed without intervention. additionally balance. mass dispensed (both liquid) gravimetrically calculate detail custom-built parts Quantos, refer S2–S4. IKA RCT digital containing mixtures. To prevent moving they stirred, holder “vision station”. open front enable monitoring contents inside (refer S6 HeinSight, open-source package previously group usually observation (Zepel 2020Zepel Lai Hein J.E. liquid-level control vision.ChemRxiv. https://doi.org/10.26434/chemrxiv.12798143.v1Crossref (0) responsible key workflow: measuring turbidity. module, specifically purpose screening, analyzes build reference library Images captured processed measures average brightness solute/solvent combination. Because optical property characterized how suspended dispersed scatter (Lawler, 2013Lawler D.M. Turbidity, Turbidimetry, Nephelometry. Reference Module Chemistry. Sciences Chemical Engineering, 2013Google organic composed near-white serves proxy relative situations. value output referenced against image pure decision made whether 2). workflow’s application meet expectations relatively bright more sophisticated exist (Liu 2018Liu Fang review vision.IEEE Access. 60586-60604Crossref (9) comparison hand-built, investment part specialized equipment. Using far simpler us validity simplest variant module. mimic would manually conduct described here script being code packages. broken stages: prepare, monitor, 3). preparation stage, sets region interest (ROI) rest input, prepares volume places station. During periodically visually monitors determines maximum safe reached. Finally, weighed properly disposed of, combination, automatically selected next Initial setup minimal. should sure clean charged bars available rate. See documentation section detail. Once started, follow programmed updating iterating through experiments combination complete interruption error pause request occurs. detailed description involved follows. run, immediately prepared: brings empty station solvent. prompted select normalize pixels monitored reference. takes tray gripping uncaps vial. cap temporarily parked pillar S5). transfers Quantos. balance tares starts z stage drives upwards, enter chamber bring back recapping. Recapping torque tighten single degree associated tightened details accessories uncapping recapping information. doses enables liquids drawn precisely there pick up uncap tooth. ensure handling done properly, smart implemented S8). order eliminate air bubbles draw liquids, first primed backing Prior drawing lines ?L gap drawn, injected bottle stored compensate drawn. remaining kept create distance line. liquid, second dripping out movements dispensed. monitored, remains fixed volumes directed These additions comprise ranging 10 30 ?L. identifies if reached, discards weigh As just measured, thus calculated directly. With determined. This discrepancies commanded dispense actual accounted calculating gathered real time. status, types, amounts errors sent Slack real-time updates slack report update provides tool especially valuable process. experiment, discussed logic duration varies depending HeinSight includes decisions input. 4 shows steps fit larger First, robot fills transferred obtain value. two ROI. ROI, normalization 2 below 5).The ROI defines area capture Yellow outlines show vessel (upper polygon) (lower triangle). proceeds steps: pre-process preselected combat issues arise environmental changes lighting, normalized stable image. recorded, stability (step 3) across consecutive images. reduces noise lighting artifacts checks trends over statistical features (mean, median, mode, deviation, range) recent measurements. If determined yet reached volume, continues. stable, comparing current “dissolved,” triggers burst agitate settled bottom stabilized minutes classified Details requirements states found illustrate applicability solvents (acetonitrile, water, acetone, ethanol, methanol). tested various challenge algorithm. 6 acetone sharp decreases correspond because stable. behavior explained incremental Following large particles stirring, manifests increased trend volumes; produced lines. trigger short stir-burst nears no left undissolved. final around 100 increase