PolyNC: a natural and chemical language model for the prediction of unified polymer properties.

Language models exhibit a profound aptitude for addressing multimodal and multidomain challenges, a competency that eludes the majority of off-the-shelf machine learning models. Consequently, language models hold great potential for comprehending the intricate interplay between material compositions and diverse properties, thereby accelerating material design, particularly in the realm of polymers. While past limitations in polymer data hindered the use of data-intensive language models, the growing availability of standardized polymer data and effective data augmentation techniques now opens doors to previously uncharted territories. Here, we present a revolutionary model to enable rapid and precise prediction of Polymer properties via the power of Natural language and Chemical language (PolyNC). To showcase the efficacy of PolyNC, we have meticulously curated a labeled prompt-structure-property corpus encompassing 22 970 polymer data points on a series of essential polymer properties. Through the use of natural language prompts, PolyNC gains a comprehensive understanding of polymer properties, while employing chemical language (SMILES) to describe polymer structures. In a unified text-to-text manner, PolyNC consistently demonstrates exceptional performance on both regression tasks (such as property prediction) and the classification task (polymer classification). Simultaneous and interactive multitask learning enables PolyNC to holistically grasp the structure-property relationships of polymers. Through a combination of experiments and characterizations, the generalization ability of PolyNC has been demonstrated, with attention analysis further indicating that PolyNC effectively learns structural information about polymers from multimodal inputs. This work provides compelling evidence of the potential for deploying end-to-end language models in polymer research, representing a significant advancement in the AI community's dedicated pursuit of advancing polymer science.

Storage Stability of 6FDA-DMB Polyamic Acid Solution Detected by Gel Permeation Chromatography Coupled with Multiple Detectors.

Polyamic acid (PAA) is the precursor of polyimide (PI), and its solution's properties have a direct influence on the final performances of PI resins, films, or fibers. The viscosity loss of a PAA solution over time is notorious. A stability evaluation and revelation of the degradation mechanism of PAA in a solution based on variations of molecular parameters other than viscosity with storage time is necessary. In this study, a PAA solution was prepared through the polycondensation of 4,4'-(hexafluoroisopropene) diphthalic anhydride (6FDA) and 4,4'-diamino-2,2'-dimethylbiphenyl (DMB) in DMAc. The stability of a PAA solution stored at different temperatures (-18, -12, 4, and 25 °C) and different concentrations (12 wt% and 0.15 wt%) was systematically investigated by measuring the molecular parameters, including Mw, Mn, Mw/Mn, Rg, and [η], using gel permeation chromatography coupled with multiple detectors (GPC-RI-MALLS-VIS) in a mobile phase 0.02 M LiBr/0.20 M HAc/DMF. The stability of PAA in a concentrated solution decreased, as shown by the reduction ratio of Mw from 0%, 7.2%, and 34.7% to 83.8% and that of Mn from 0%, 4.7%, and 30.0% to 82.4% with an increase of temperature from -18, -12, and 4 to 25 °C, respectively, after storage for 139 days. The hydrolysis of PAA in a concentrated solution was accelerated at high temperatures. Notably, at 25 °C, the diluted solution was much less stable than the concentrated one and exhibited an almost linear degradation rate within 10 h. The Mw and Mn decreased rapidly by 52.8% and 48.7%, respectively, within 10 h. Such faster degradation was caused by a greater water ratio and less entanglement of chains in the diluted solution. The degradation of (6FDA-DMB) PAA in this study did not follow the chain length equilibration mechanism reported in literature, given that both Mw and Mn declined simultaneously during storage.

Influence of Heating Rate on the Structure and Mechanical Properties of Aromatic BPDA-PDA Polyimide Fiber.

Aromatic polyimide fibers (PI) are usually produced in two steps. The precursor fibers of polyamic acid (PAA) are fabricated first, and then the fabricated fibers are converted into PI fibers through thermal treatment. In the second step (thermal treatment), the mechanical properties of the obtained PI fibers are remarkably affected. Here, the PAA fibers derived from 3,3',4,4'-biphenyltetra-carboxylic dianhydride and p-phenylenediamine are fabricated by a dry-jet wet-spinning method. Then, the PI fibers are prepared by heating PAA fibers from room temperature to 300, 350 and 400 °C under different heating rates, ranging from 1 °C/min to 80 °C/min. When the heating rate is low, the crystallization lags behind the imidization process, and begins only when the imidization degree reaches a high level. As the heating rate increases, the crystallization tends to occur simultaneously with the imidization process, and the degree of crystallinity of the PI fibers also greatly increases. Our findings suggest that a high heating rate causes the polymer chains to undergo high mobility during thermal treatment. The tensile modulus of the PI fiber further demonstrates a high dependence on the heating rate. Moreover, a short annealing process after treatment proves to be efficient in releasing residual stress and improving tensile strength.

Associative behavior of polyimide/cyclohexanone solutions.

Our previous work has demonstrated that soluble polyimide with relatively weak interaction can be transformed from neutral polymer to associative polymer by increasing molecular weight. Thus, it is necessary to find another way to vary the relatively weak interaction strength, i.e. variation of solvent quality. Herein, viscoelastic behaviors are examined for 2,2-bis(3,4-dicarboxy-phenyl) hexafluoropropane dianhydride (6FDA)-2,2'-bis(trifluoromethyl)-4,4'-diam (TFDB) polyimide (PI), with a relatively low molecular weight (M w) of 88 000 g mol-1, dissolved in cyclohexanone (CYC). The scaling relationship between viscosity (η 0-η s) and volume fraction is in good agreement with the associative polymer theory proposed by Rubinstein and Semenov. Oscillatory rheological results indicate that the PI solution tends to become a gel with increased volume fraction. The synchrotron radiation small-angle X-ray scattering results imply the existence of dense aggregates in the concentrated PI/CYC solutions. Shear thickening and thinning behaviors are observed in the solutions, and the shear thickening behavior of polyimide solution has not been reported in literature. Their mechanisms are studied by conducting dynamic and steady rheological experiments. Thus, enhancing the relatively weak interaction strength can also make the low M w polyimide show associative polymer behavior. This work can help us to gain deep insight into polyimide solution properties from dilute to semidilute entangled solutions, and will guide the preparation of polyimide solutions for different processing.

Viscoelastic behaviour and relaxation modes of one polyamic acid organogel studied by rheometers and dynamic light scattering.

A novel polyamic acid (PAA from BAPMPO-BPDA) organogel was synthesized and characterized via dynamic light scattering (DLS), a classical rheometer, and diffusion wave spectroscopy (DWS). In situ monitoring was performed using a classical rheometer to observe the formation of the PAA organogel. The rheological curves confirm the formation of the PAA gel network and the origin of hydrogen bonding from the -NH- group (donor) and P[double bond, length as m-dash]O group (acceptor). The autocorrelation functions of PAA under different conditions (pure gel, gel with NaNO3, gel with formamide) are measured via DLS, and different characteristic times are obtained via the CONTIN method. Three different relaxation modes of the PAA gel, i.e., fast, intermediate and slow modes, are observed. The fast and intermediate modes show a diffusive behaviour (τ ∼ q-2), whereas the slow mode did not. When enough formamide is added into the PAA gel, the fast mode disappears; addition of enough salt (NaNO3) leads to disappearance of the slow mode. The relationship between characteristic time and diffusion vector demonstrates that the different decorrelation modes consisted of two homodyne and two heterodyne components. Two single-exponential functions and two stretched exponential functions were used, and the different decorrelation modes of the PAA gel are expressed with a non-linear function, which fits the autocorrelation function very well. And the different decorrelation modes are also discussed. DWS results in the high-frequency region not only demonstrate the formation of a PAA gel network but also indicate that the semiflexible chains of PAA are due to electrostatic interaction. The DWS results at different time scales are analyzed by applying the de Gennes' reptation model.

A weighted multipath measurement based on gene ontology for estimating gene products similarity.

Many different methods have been proposed for calculating the semantic similarity of term pairs based on gene ontology (GO). Most existing methods are based on information content (IC), and the methods based on IC are used more commonly than those based on the structure of GO. However, most IC-based methods not only fail to handle identical annotations but also show a strong bias toward well-annotated proteins. We propose a new method called weighted multipath measurement (WMM) for estimating the semantic similarity of gene products based on the structure of the GO. We not only considered the contribution of every path between two GO terms but also took the depth of the lowest common ancestors into account. We assigned different weights for different kinds of edges in GO graph. The similarity values calculated by WMM can be reused because they are only relative to the characteristics of GO terms. Experimental results showed that the similarity values obtained by WMM have a higher accuracy. We compared the performance of WMM with that of other methods using GO data and gene annotation datasets for yeast and humans downloaded from the GO database. We found that WMM is more suited for prediction of gene function than most existing IC-based methods and that it can distinguish proteins with identical annotations (two proteins are annotated with the same terms) from each other.