Jules
Lee
a,
Prajakatta
Mulay
a,
Matthew J.
Tamasi
a,
Jonathan
Yeow
b,
Molly M.
Stevens
bc and
Adam J.
Gormley
*a
aDepartment of Biomedical Engineering, Rutgers, The State University of New Jersey, Piscataway, NJ 08854, USA. E-mail: adam.gormley@rutgers.edu
bDepartment of Materials, Department of Bioengineering, and Institute of Biomedical Engineering, Imperial College London, London, SW7 2AZ, UK
cDepartment of Medical Biochemistry and Biophysics, Karolinska Institutet, SE-171 77 Stockholm, Sweden
First published on 5th January 2023
Oxygen tolerant polymerizations including Photoinduced Electron/Energy Transfer-Reversible Addition–Fragmentation Chain-Transfer (PET-RAFT) polymerization allow for high-throughput synthesis of diverse polymer architectures on the benchtop in parallel. Recent developments have further increased throughput using liquid handling robotics to automate reagent handling and dispensing into well plates thus enabling the combinatorial synthesis of large polymer libraries. Although liquid handling robotics can enable automated polymer reagent dispensing in well plates, photoinitiation and reaction monitoring require automation to provide a platform that enables the reliable and robust synthesis of various polymer compositions in high-throughput where polymers with desired molecular weights and low dispersity are obtained. Here, we describe the development of a robotic platform to fully automate PET-RAFT polymerizations and provide individual control of reactions performed in well plates. On our platform, reagents are automatically dispensed in well plates, photoinitiated in individual wells with a custom-designed lightbox until the polymerizations are complete, and monitored online in real-time by tracking fluorescence intensities on a fluorescence plate reader, with well plate transfers between instruments occurring via a robotic arm. We found that this platform enabled robust parallel polymer synthesis of both acrylate and acrylamide homopolymers and copolymers, with high monomer conversions and low dispersity. The successful polymerizations obtained on this platform make it an efficient tool for combinatorial polymer chemistry. In addition, with the inclusion of machine learning protocols to help navigate the polymer space towards specific properties of interest, this robotic platform can ultimately become a self-driving lab that can dispense, synthesize, and monitor large polymer libraries.
Controlled/Living Radical Polymerization (CLRP) via Reversible Addition–Fragmentation Chain Transfer (RAFT) polymerization has been previously used to generate libraries of new and innovative polymers in parallel for biomedical applications.8–15 Recently, multiple research groups have developed impressive automated platforms specifically for high-throughput polymer synthesis. For example, Lin et al. reported a platform technology that allowed for the generation of a two-dimensional library of 100 diblock polyester and polycarbonate copolymers in under nine minutes using continuous-flow reactors.16 Additionally, Leibfarth and co-workers developed a flow system that accelerated the synthesis of sequence-defined copolymers by semiautomated iterative exponential growth.17 Chen and co-workers used a droplet flow system to generate 11 statistical copolymers within 11 minutes.18 However, the reaction monitoring must be performed manually, which slows down the overall development of novel polymers. Therefore, few groups have further improved these platform technologies by including automated characterization methods into the workflow. For example, Warren and Junkers demonstrated operator-independent online monitoring of RAFT polymerizations via flow chemistry using an inline benchtop NMR.19,20 Meanwhile, Lauterbach et al. reported the use of an inline UV-Vis spectrophotometer to monitor the flow polymerization online.21 Junkers recently incorporated inline size-exclusion chromatography (SEC) and NMR, which is a promising platform that may also enable high-throughput parallel synthesis, kinetic screenings, and robust characterization of polymers.22 While such automated platforms provide great utility for flow-based RAFT polymerization methods, there is an unmet need for the development of a platform that generates polymer libraries in a batch process with real-time monitoring of reaction conversions.
The discovery of oxygen-tolerant Photoinduced Electron/Energy Transfer-Reversible Addition–Fragmentation Chain Transfer (PET-RAFT) polymerization has enabled polymerizations to be performed in a well plate format on the benchtop.23–26 Boyer and co-workers used a photo-initiator that served a dual purpose of initiating the polymerization as well as aiding the singlet oxygen trapping for facilitating oxygen tolerance, whereas Gibson and co-workers achieved oxygen tolerance using triethanolamine as the degassing agent.24,27–29 This oxygen tolerance has further enabled the development of high-throughput polymer chemistry.30 However, creating polymer libraries in well plate formats requires manually adding the reagents in individual wells which is time-consuming and low-throughput (Fig. 1A-1). Our group recently demonstrated the utilization of liquid-handling robotics with PET-RAFT polymerizations that allowed for high-throughput synthesis of combinatorial polymer libraries in well plate format (Fig. 1B-1 and B-2).4 This liquid handling system reduces >80% of the time required for reagent dispensing and allows for a fully automated platform for combinatorial polymer chemistry. However, the use of high-throughput photochemistry in well plates requires additional challenges to be overcome to achieve a fully robotic platform.
The first challenge for end-to-end automated PET-RAFT polymerizations is providing automated control over the photoinitiation of individual reactions in each well. Previous experiments have demonstrated that the initiation of a PET-RAFT reaction is solely dependent on exposure to light, which provides control over polymerization initiation and termination.31,32 Conventional methods for PET-RAFT polymerization require that the well plate containing the reaction mixture be placed directly under an LED lamp or an array.24,31,33,34 However, these methods are limited in performing combinatorial PET-RAFT synthesis of various polymer compositions since they do not offer control over individual reactions in each well (Fig. 1A-2). For example, excess light exposure in one well may result in continuous RAFT activation in the absence of monomer leading to undesirable side reactions such as loss of RAFT end group functionality or even polymer coupling events that would skew the desired polymeric attributes such as the molecular weight and dispersity. A solution to this limitation is the design of a device with addressable LEDs in which the lighting of individual wells is controlled to provide spatial and temporal control of PET-RAFT reactions and enable combinatorial polymer chemistry in which each polymer attains the desired molecular weight and dispersity. 96-LED lightboxes are also used in biological research.35,36 For example, Amuza Inc provides a commercially available 96-LED matrix for optogenetic applications, while Henkel's LOCTITE® LED flood system uses an array of UV or visible light LEDs for adhesive curing.37 However, these lightbox designs are limited in their applications for high-throughput PET-RAFT polymerization since there is (i) no control over individual LEDs, (ii) limited compatibility with automation, (iii) are expensive costing over $6000, and (iv) are not available in multiple wavelengths. Thus, a lightbox with individually addressable LEDs that can independently provide optimal light exposure for robust and reliable high-throughput combinatorial well plate PET-RAFT polymerizations is desired to enable a highly precise level of control over polymerization.
Significant potential also exists to further automate PET-RAFT polymerizations and improve control over the polymerization process by integrating techniques for in situ, continuous monitoring of the reaction progress. Yeow et al. have recently shown that using 5,10,15,20-tetraphenyl-21H,23H-porphine zinc (ZnTPP) as a photocatalyst for PET-RAFT reactions allows for online monitoring of the reaction by tracking changes in the photocatalyst fluorescence.32 They observed that the evolution of the ratio of the emission intensities at 632 and 615 nm (I632/I615) was highly correlated to the evolution of monomer conversion and thus allowing PET-RAFT polymerizations to be monitored in real-time.
Here, we report a novel platform technology consisting of a custom-designed lightbox, a robotic arm, and a fluorescence plate reader (Fig. 1B-3) for automating PET-RAFT polymerizations in well plates to obtain a library of polymers in parallel. The custom-built lightbox is able to individually address LEDs that are automatically controlled throughout the polymerization using Python. It can be easily fabricated at a low cost of approximately $100 and allows for customizations for specific purposes depending on the user. In addition, these polymerizations are tracked via online fluorescence monitoring on the spectrophotometer without manual intervention. Before initiating the polymerization with the lightbox, fluorescence intensities are obtained, and the fluorescence data is then imported to Python where fluorescence ratios are calculated and plotted. The data is then sent to the lightbox to control the individual LEDs. The robotic arm then performs a plate transfer from the fluorescence plate reader to the lightbox to initiate PET-RAFT polymerizations, then transfers back to the fluorescence plate reader for another fluorescence read after 30 minutes of photopolymerization. This process is repeated until Python calculates that the slopes between the two most recent consecutive fluorescence data time points are below a set threshold for each reaction in the well plate and then deems the polymerizations to be complete. The LEDs are then automatically turned off for the completed polymerizations by Python. Thus, this robotic platform allows for fully automated online fluorescence tracking of PET-RAFT polymerizations and enables a method to provide an extremely high level of spatial and temporal control of individual PET-RAFT reactions in a well plate format. Finally, building upon our previous work, we also report the facile integration of Hamilton Microlab STARlet liquid handling robotics4 with our robotic platform to enable high-throughput polymerizations of acrylate homopolymers and copolymers synthesized in parallel.
The saved XML file was then imported to Python to be analyzed for lightbox control. Empty wells that showed negligible fluorescence values were replaced with a “0” value, and the fluorescence ratios for wells containing the reaction mixture were calculated using eqn (1),
(1) |
This 8 × 12 matrix for 96 wells was saved internally to later display a list of fluorescence ratios for wells containing PET-RAFT reactions. A copy of the fluorescence data matrix was further manipulated in Python to ultimately control LED lighting. The fluorescence data matrix was converted into a binary logic array, Aij, where non-zero fluorescence values were replaced with a “1” value, while the “0” values remained. Aij was then multiplied by matrix Bj shown in eqn (2),
Bj = 2j | (2) |
(3) |
After the initial fluorescence read, a Hudson Robotics PlateCrane EX robotic arm was used to perform plate transfers between the fluorescence plate reader and the lightbox. PET-RAFT polymerizations were allowed to proceed for 30 minutes and were then transferred from the lightbox back to the fluorescence plate reader by the robotic arm. The fluorescence read, plate transfer, and lighting calculations previously described were repeated, and the slope of the line joining the two consecutive fluorescence ratios was determined using eqn (4),
(4) |
If the calculated slope at the two most recent time points of a reaction well was lower than a set threshold, as indicated by the plateauing of the slope, the polymerization reaction in the well was deemed complete and the corresponding LED was turned off. The automation of the polymerization reactions continued until the obtained slope for each reaction well was below the set threshold.
Fig. 3 Heatmap of light intensities at different well positions using (left) lightbox with all LEDs on (variance = 2.3%) and (right) an overhead lamp (variance = 29.2%). |
To validate the hypothesis that excess photoinitiation in a PET-RAFT polymerization could lead to over polymerization, two replicates of pDMA (DP 200) and pMEA (DP 200) were polymerized for three hours and 24 hours in the center wells of a polystyrene 96 well-plate. Comparison of SEC traces demonstrated that prolonged polymerization times led to the presence of distinct high molecular weight shoulders for both monomers (Fig. S3†). We hypothesized that the prolonged irradiation led to dimerization of the polymer chains due to continual activation of the RAFT end group. These side reactions are undesirable not only for the increase in polymer dispersity but also for the loss of end chain fidelity (livingness) which would severely limit additional chain extension of the polymer or the efficiency of post-modification reactions. They also show the formation of higher molecular weight populations indicated by the bimodal trace of pDMA and a shoulder in pMEA irradiated for 24 hours. The formation of these higher molecular weight products caused by the dimerization of polymer chains due to excess photoinitiation is undesirable as they negatively affect the dispersity of the polymer. In addition, these side products may affect the end chain fidelity and livingness of the reaction and hamper additional chain extension of the polymer in case of synthesis of block copolymers. Over-polymerization may deter performing combinatorial chemistries of different monomers with different reactivities throughout a single plate. Thus, a lightbox with individually addressable LEDs that controls the radiation times specifically tailored to each PET-RAFT polymerization in a well plate could ensure that high monomer conversions are reached while preventing the negative effects of over polymerization.
To further investigate the applicability of our custom-built lightbox for PET-RAFT polymerizations, two replicates of pDMA (DP 200) were synthesized in wells A1 and A12 of a 96-well plate, and their respective LEDs were periodically turned on and off every 30 minutes by the automated platform where the lighting of wells A12 and A1 alternated after every time point. Fluorescence tracking was performed at every timepoint to monitor the reaction progress of each well (Fig. 2C). The fluorescence ratio for well A1 increased in the first 30 minutes when its corresponding LED was on, however, the fluorescence ratio remained constant, and the monomer conversion ceased for the next 30 minutes when its LED was off. The fluorescence ratio increased again after the corresponding LED was turned on suggesting that the livingness of the polymerization and the photoactivity of ZnTPP were conserved. The response of the fluorescence ratios to the status of the LED was consistent throughout the experiment. Concurrently, a similar response was obtained for the polymerization in well A12. These results indicated that the inherent activation and inactivation character of PET-RAFT mediated by light exposure could be tightly regulated at the individual well level in well plates by the implementation of individually addressable LEDs. In contrast to previous work with PET-RAFT polymerization whereby a single light source was used to control the polymerization process of the entire well plate, our lightbox enabled additional dimensions of PET-RAFT control at the spatial and temporal levels in single wells of a 96-well plate.32
The trend of the evolution of fluorescence ratios at different time points demonstrated by the lightbox and the lamp show similarity across all monomers except for pDMA, which displayed lower fluorescence ratio magnitudes on the lightbox than on the lamp (Fig. 4B). However, higher monomer conversions (calculated through 1H-NMR spectroscopy) were obtained with lightbox than lamp at all time points. All polymers synthesized on the lightbox demonstrated more than 90% final conversion except for pEA and pDMA, which reached a final conversion of 85% and 75%, respectively, and all polymers synthesized on the lamp showed final conversions ranging between 68% and 80%. The higher conversions observed using the lightbox may have been the result of the larger spectral profile of the lightbox compared to the lamp leading to an increased activation of ZnTPP when using warm-white LEDs as the photoinitiation source.39,40 Shanmugam et al. have reported that multiple color wavelengths can activate ZnTPP at different rates, where yellow, green, and orange light cause fast polymerization while colors such as red and blue result in relatively slower polymerizations.40 The broad absorbance of ZnTPP may therefore have caused increased ZnTPP activation and faster polymer reaction kinetics on the lightbox compared to the lamp. Interestingly, all homopolymers displayed linear correlations of the fluorescence ratio with monomer conversion consistent with the findings of Yeow et al. (Fig. 4C and D).32 The calculation of adjusted R2 linear correlation coefficients of all homopolymers indicated a high correlation between fluorescence ratios and conversions for all homopolymers regardless of photoinitiation source (Table S2†). pMA synthesized on the lightbox and lamp displayed correlation coefficients of 0.899 and 0.832, respectively, while all other homopolymers displayed correlation coefficients >0.95. Analysis of the correlation curves between homopolymers synthesized on the lightbox and those synthesized on the lamp show that although there was some variation in the fluorescence ratios and their associated conversions for each homopolymer, the lightbox was still able to synthesize PET-RAFT polymers similar to that of the lamp. The observation of high monomer conversions and strong linear correlations between fluorescence ratio and conversion for polymers synthesized on the lightbox provided strong evidence that our custom-designed lightbox could enable robust PET-RAFT reactions in parallel. In addition, we hypothesized that the use of automated LED multiplexing could allow for control of individual PET-RAFT reactions by using the slopes between fluorescence ratios calculated between consecutive time points to monitor reaction progress instead of the fluorescence ratio values as an indication of reaction termination.
We then investigated the utility of our automated platform and the impact of multiplexed lighting controlled by slopes of fluorescence ratios obtained between consecutive time points in each well on polymer attributes. A slope threshold of 0.002 was set in Python to ensure that all polymers synthesized received maximum light exposure and achieved high monomer conversions. Parallel PET-RAFT polymerization reactions were performed where reagents for pHEA, pDMA, pMEA, pEA, pMA, and pNAM were dispensed in the first column of a 96-well plate manually and were polymerized on our automated platform with feedback-controlled lighting. The slopes for fluorescence ratios vs. time plots of all acrylate monomers polymerized with the lightbox plateaued at t = 2 h time points, and thus their corresponding LEDs were switched off automatically, while NAM and DMA were deemed to be completely polymerized at t = 2.5 h by our platform (Fig. 5A). This difference in plateauing at different fluorescence ratio values may have been caused by the differences in ZnTPP (photocatalyst) incorporation into the different polymer backbones, as suggested by Yeow et al.32
The conversions of these homopolymerizations were determined using 1H-NMR spectroscopy. The conversion kinetics of the homopolymers that underwent LED multiplexed lighting were similar to those of their respective homopolymers synthesized using constant lightbox radiation (Fig. S4†). Final conversions obtained at 2.5 h for all polymers again showed high final conversions of above 85%, demonstrating that the use of multiplexed lighting provided automatic control of each polymerization while high final conversions were achieved. Linear correlations between the fluorescence ratios and the observed conversion at each time point calculated through 1H-NMR spectroscopy of homopolymers synthesized using feedback-controlled LED lighting again showed strong correlations, as R2 values for all polymers were >0.95 except for pNAM which displayed a lower R2 value of 0.89 (Fig. 5B, Table S3†). Since NAM demonstrated more than 85% conversion within the first time point of 30 minutes, we tracked NAM conversion on the lightbox with increased time resolution of 5 minutes (Fig. S5†) for optimizing the linear correlation fit. Similar fitting equations were obtained with both the time resolutions, that is y = 0.00625x + 0.57797 with R2 of 0.96 for a time resolution of 5 minutes as compared to y = 0.0071x + 0.54747 with R2 of 0.95 obtained with 30 minute time resolution, thus representing the true fit. In addition, an increase in the fluorescence ratio was observed at constant NAM conversion beyond 90 minutes leading to a poor fit. This increase may have been caused by additional fluorescence changes induced by the non-linear incorporation of ZnTPP into the polymer backbone at very high NAM conversions. Since the slope threshold of 0.002 for automated light cutoffs may have been too low for NAM polymerization, it over-exposed the NAM polymerization at high conversion that induced a non-linear incorporation of ZnTPP in the polymer backbone. To optimize the slope threshold for NAM polymerization, its conversion was tracked at a resolution of 15 minutes for 2.5 h (Fig. S6†). More than 95% conversion was obtained within 90 minutes where the slope of 0.00284 was still above the set threshold of 0.002 that continued the reaction further and thus increasing the fluorescence ratio. Therefore, for highly reactive species such as NAM, setting a higher threshold of 0.003 is suggested. This indicated that determining fluorescence ratios to provide tailored lighting times to each polymer reaction can provide correlations to the actual monomeric conversions when performing PET-RAFT polymerizations using either the lightbox or the lamp as the light source, but direct correlations between the fluorescence ratios and experimental conversions are dependent on the specific monomer used. In addition, these results indicate that the use of slope calculations between fluorescence ratios provides a method that enables each PET-RAFT reaction within the separate wells of a well plate to receive the optimal amount of light exposure to ensure full conversion and potentially minimize the effects of over-polymerization. Although there may be differences in final fluorescence ratio values, this system takes advantage of the fluorescence ratio plateauing behavior conserved throughout all PET-RAFT reactions to provide flexible and robust synthesis of different PET-RAFT polymers in parallel. Finally, we anticipate that the use of this fluorescence monitoring method will also be of particular utility in quality control processes to determine the progression of each PET-RAFT reaction.
To confirm polymerization was controlled according to a PET-RAFT mechanism, polymers were analyzed with SEC (Fig. S7, Table S4†). Comparisons between the SEC traces of homopolymers synthesized on the lamp versus the lightbox show that polymers synthesized on the lightbox showed similar molecular weights and molecular weight distributions thereby demonstrating our platform's ability to provide control over individual polymerization reactions. It should be noted that the molecular weights obtained for HEA homopolymers were higher than the theoretical molecular weight which was attributed to the difference in hydrodynamic diameter between pHEA and the PMMA standards. Homopolymers synthesized on the lightbox also demonstrated similar polymer dispersities compared to those obtained with the lamp. Overall, our custom-built lightbox demonstrated success in performing PET-RAFT polymerizations comparable to those synthesized using a conventional lamp-based illumination source and provided the advantage of automated individual well control for parallel polymer synthesis to ensure intermediate to high conversion while providing a mechanism for the prevention of over polymerization.
Copolymers synthesized using our automated robotic platform demonstrated fluorescence ratios similar to those of the homopolymers, where the evolution of fluorescence ratio for each polymer with time displayed a logarithmic trend (Fig. 7A–D). The slopes for fluorescence ratios vs. time plots of all the copolymers containing acrylate monomers plateau between t = 1 h and t = 2 h, and therefore the reaction was deemed complete, and their LEDs were turned off automatically. All copolymers containing exclusively acrylamides were completed after t = 3 h. CP1, CP2, CP7, CP10, and CP11 copolymers with HEA-co-NAM, MEA-co-MA, EA-co-HEA-co-MA, DMA-co-HEA-co-MEA-co-EA, and MEA-co-NAM-co-EA-co-DMA compositions were synthesized by 1.5 h; and CP3, CP5, CP6, and CP9 with DMA-co-EA, HEA-co-MA-co-NAM, MEA-co-EA-co-DMA, and NAM-co-MA-co-HEA-co-MEA compositions were completely polymerized by 2 h. For homopolymer reactions, EA and MA were deemed to be completely polymerized at 1.5 h, HEA at 2 h, DMA at 2.5 h, and NAM and MEA at 3 h. In addition, copolymers containing only acrylamide monomers: NAM-co-DMA and NAM-co-DMA-co-NIPAM, displayed the same behavior as their homopolymer counterparts in which they reached high fluorescence ratios even after full conversion, yet this behavior was prevented with the incorporation of an acrylate monomer into the copolymer composition. Despite the variation in fluorescence ratios, all homopolymers and copolymers showed intermediate to high conversions, where total conversions of homopolymers ranged from 75 to 94%, copolymers with two monomer compositions ranged from 75 to 98%, copolymers with three monomer compositions ranged from 79 to 90%, and copolymers with four monomer compositions ranged from 88 to 92% (Fig. 7E) as calculated using 1H-NMR spectroscopy. Intermediate to high conversions were obtained for each monomer independent of desired polymer composition, which showed the versatility of using fluorescence tracking for synthesizing acrylate/acrylamide homopolymers and copolymers. SEC traces further demonstrated desired molecular weights (Fig. 7A–D, Table S5†), however, the molecular weights obtained for copolymers containing HEA were higher than the theoretical molecular weight due to the swelling of HEA in solution. It was expected that the Đ of the copolymers to be slightly higher than the homopolymers due to the different reactivities between different monomer families in solution, but good control over the polymerization using our automated platform was demonstrated with relatively low dispersities throughout. Synthesis of both homopolymers and copolymers with control of each well using our automated platform resulted in polymer synthesis with high final conversions and good agreement between the theoretical and observed molecular weights while obtaining low dispersities. In addition, the use of slope calculations to determine the completion of reactions proved to be flexible enough to provide robust polymer synthesis regardless of monomer composition and provided strong evidence that our automated platform can be a powerful tool for high-throughput combinatorial chemistry in well plates.
This automated platform could be further challenged to synthesize larger polymer libraries that consist of methacrylate and methacrylamide monomers, as well as different polymer architectures such as block copolymers and star copolymers. Methacrylate and methacrylamide monomers demonstrated slow changes in fluorescence intensity ratio lower than the set threshold of 0.002 between 30 minute time points which caused early termination of reaction by Python (not reported), and therefore, optimization of time points is required for each family of monomers. This caveat as well as the fact that a slope threshold of 0.002 seems to be too low for acrylamide homopolymers underlines how our automated platform could be vastly improved by assigning a slope threshold to each individual reaction based on which monomer species is being used. The versatility of this automated platform can be further improved by the use of specifying lighting patterns for each timepoint using the experimental template by entering the binary data array to provide control over lighting times in each well. The creation of this automated platform also opens the door to the inclusion of different automated workflows and assays. For example, the addition of another lightbox with UV LEDs would allow for automated side chain functionalization of polymer backbones synthesized on our platform and allow for increased exploration into polymer structure–function relationships.41
Our automated platform can be further expanded by integration of additional instruments, as well as the addition of automated data analysis workflows. For example, analytical techniques essential for polymer characterization such as SEC could be optimized by automating processes for sample preparation and trace analysis using the liquid handler and Python, respectively. Furthermore, the use of high-throughput SEC columns allowing run times of four minutes per sample at the sacrifice of resolution would significantly improve analytical capacity. In addition, the inclusion of more experimental steps in Python could allow for automated enzyme kinetics assays, and the addition of a dynamic light scattering plate reader would allow for high-throughput hydrodynamic radius screening of different polymer compositions. Finally, further development of this platform with the inclusion of machine learning protocols would enable decisive and effective navigation of the highly multidimensional space that different polymer compositions and architectures provide. With the large amounts of data that automation can provide in the field of polymer chemistry, multiple groups have already shown how machine learning can take full advantage of these large datasets. Warren et al. were able to feed data generated by an online NMR and inline GPC into a closed-loop machine learning algorithm to optimize polymerization reaction conditions and directly influence polymer characteristics such as dispersity and conversion.42 Similarly, Leibfarth et al. demonstrated the use of active machine learning combined with automated polymer synthesis to identify effective 19F MRI copolymer agents.43 Hartman et al. utilized an artificial neural network to predict metallocene-catalyzed olefin polymerizations.44 In our own work, the inclusion of active learning with our automated platform informs polymer synthesis and characterization until the desired polymer structure–function behavior is found, such as polymers that self-assemble into single-chain polymer nanoparticles or polymers that can form polymer-protein hybrids.6,7,45 This self-driving laboratory in which a design-build-test-learn paradigm is employed to systematically discover new polymers with specific activities will allow nonexperts to create highly complex polymer designs reliably and efficiently.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d2dd00100d |
This journal is © The Royal Society of Chemistry 2023 |