Antibiotic resistance has emerged as a major public health threat globally in both developing as well as the developed countries and without multi pronged interventions its status is going to worsen in future as per World Health Organization, (http://www.who.int/mediacentre/factsheets/antibiotic-resistance/en/). Leading cause of drug resistance is the growth of microorganisms in biofilm mode. Biofilms are colonized behaviour of microbes, where a unicellular bacteria mimics multicellular lifestyle by undergoing division of labour. Planktonic bacteria go through various modifications to withstand unfavourable conditions (like antibiotics) by transforming in a self-secreted cocoon called biofilm. The biofilms are evident to be 10-1000 folds more resistant towards antibiotics. Anti-biofilm agents can disrupt the biofilms and can also enhance the conventional antibiotics through synergistic effects similar to adjuvants increasing the efficacy of vaccines. Anti-biofilm agents even demonstrated greater promises by killing the Multidrug-resistant strains including ESKAPE pathogens (Carol Potera 2010).

Researchers have been working hard to develop various anti-biofilm agents from last three decades due to their immense therapeutic potential. However computational resources in this important field are absolutely lacking. To fill this void, we have recently developed a resource of anti-biofilm agents called “aBiofilm” that harbors 1720 unique anti-biofilm agents among 5027 entries (Rajput A et al 2018) targeting 140 micro-organisms. It also provides a simplified classification based prediction tool for identification of anti-biofilm agent. However, there is an urgent need to develop comprehensive web server to predict the anti-biofilm activity of chemical(s) and peptides to speed up the research in the area.

In this endeavour, we have now developed Biofilm-i, a platform to predict the biofilm inhibiting efficacy of chemical(s) and peptide(s) using Quantitative structure–activity relationship (QSAR) based regression models.

Following are the salient features of our web server:

• All regression based predictive models were developed using the curated data from aBiofilm resource (NAR database 2018)


• We have developed eight QSAR based predictive models (01 general chemical, 03 group-specific, 04 species specific)


• Data sets include training/testing and independent validation i) Chemicals = 884 unique (T ⁸⁰⁰ + V ⁸⁴) ii) Gram-positive bacteria = 384 unique (T ³⁵⁰ + V ³⁴) iii) Gram-negative bacteria = 498 unique (T ⁴⁵⁰ + V ⁴⁸) iv) Fungus or Yeast =158 unique (T ¹⁴⁰ + V ¹⁸) v) Pseudomonas aeruginosa = 301 unique (T ²⁷⁰ + V ³¹) vi) Staphylococcus aureus = 239 unique (T ²¹⁰ + V ²⁹) vii) Candida albicans = 152 unique (T ¹⁴⁰ + V ¹²) viii) Escherichia coli = 103 unique (T ⁹³ + V ¹⁰)


• Firstly we calculated the chemical descriptors (2D, 3D and fingerprints) using PaDEL software. Secondly, we extracted the best contributing features among all the chemical descriptors using “Remove Useless” and “Best First ” algorithm individually. Thirdly, the top-most features were utilized for model development through Support Vector Machine


• These QSAR methods performed with Pearson correlation coefficient (PCC) of 0.60, 0.77, 0.62, 0.77, 0.73, 0.83, 0.70, and 0.71 during 10-fold cross validation using support vector machine (SVM)


• Furthermore, all the models performed equally well with PCC of 0.59, 0.72, 0.85, 0.82, 0.89, 0.61 on the independent validation datasets


• Chemical format conversion from structure to sdf as well as draw JSME tool and analogue generator tool etc are also provided


• Chemical analyses tree map, cloud, scaffold tree are also given to display the chemical diversity of the data used

Biofilm-i, is the first regression based portal for the prediction of biofilm inhibition of chemical(s). This web server will help the wider scientific community to speed up research to tackle the menace of Antibiotic drug resistance.